Commit 6e3cf8b0 authored by Jing Zhang's avatar Jing Zhang
Browse files

merge develop

parents 4ad62d7f ba58a93f
...@@ -119,7 +119,7 @@ template <int NDimSpatial, ...@@ -119,7 +119,7 @@ template <int NDimSpatial,
void profile_conv_fwd_bias_relu_atomic_add_impl(int do_verification, void profile_conv_fwd_bias_relu_atomic_add_impl(int do_verification,
int init_method, int init_method,
bool do_log, bool do_log,
int nrepeat, bool time_kernel,
ck::index_t N, ck::index_t N,
ck::index_t K, ck::index_t K,
ck::index_t C, ck::index_t C,
...@@ -275,7 +275,8 @@ void profile_conv_fwd_bias_relu_atomic_add_impl(int do_verification, ...@@ -275,7 +275,8 @@ void profile_conv_fwd_bias_relu_atomic_add_impl(int do_verification,
{ {
std::string conv_name = op_ptr->GetTypeString(); std::string conv_name = op_ptr->GetTypeString();
float ave_time = invoker_ptr->Run(argument_ptr.get(), nrepeat); float ave_time =
invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
std::size_t flop = std::size_t(2) * N * K * Ho * Wo * C * Y * X; std::size_t flop = std::size_t(2) * N * K * Ho * Wo * C * Y * X;
......
...@@ -41,7 +41,7 @@ template <int NDimSpatial, ...@@ -41,7 +41,7 @@ template <int NDimSpatial,
void profile_conv_fwd_bias_relu_impl(int do_verification, void profile_conv_fwd_bias_relu_impl(int do_verification,
int init_method, int init_method,
bool do_log, bool do_log,
int nrepeat, bool time_kernel,
ck::index_t N, ck::index_t N,
ck::index_t K, ck::index_t K,
ck::index_t C, ck::index_t C,
...@@ -207,7 +207,8 @@ void profile_conv_fwd_bias_relu_impl(int do_verification, ...@@ -207,7 +207,8 @@ void profile_conv_fwd_bias_relu_impl(int do_verification,
{ {
std::string conv_name = op_ptr->GetTypeString(); std::string conv_name = op_ptr->GetTypeString();
float ave_time = invoker_ptr->Run(argument_ptr.get(), nrepeat); float ave_time =
invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
std::size_t flop = std::size_t(2) * N * K * Ho * Wo * C * Y * X; std::size_t flop = std::size_t(2) * N * K * Ho * Wo * C * Y * X;
......
...@@ -269,7 +269,7 @@ template <int NDimSpatial, ...@@ -269,7 +269,7 @@ template <int NDimSpatial,
bool profile_convnd_bwd_data_impl(int do_verification, bool profile_convnd_bwd_data_impl(int do_verification,
int init_method, int init_method,
bool do_log, bool do_log,
int nrepeat, bool time_kernel,
ck::index_t N, ck::index_t N,
ck::index_t K, ck::index_t K,
ck::index_t C, ck::index_t C,
...@@ -410,7 +410,8 @@ bool profile_convnd_bwd_data_impl(int do_verification, ...@@ -410,7 +410,8 @@ bool profile_convnd_bwd_data_impl(int do_verification,
{ {
std::string conv_name = conv_ptr->GetTypeString(); std::string conv_name = conv_ptr->GetTypeString();
float ave_time = invoker_ptr->Run(argument_ptr.get(), nrepeat); float ave_time =
invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
std::size_t flop = std::size_t flop =
ck::utils::conv::get_flops(N, C, K, filter_spatial_lengths, output_spatial_lengths); ck::utils::conv::get_flops(N, C, K, filter_spatial_lengths, output_spatial_lengths);
......
...@@ -65,7 +65,7 @@ template <typename ADataType, ...@@ -65,7 +65,7 @@ template <typename ADataType,
void profile_gemm_bias_2d_impl(int do_verification, void profile_gemm_bias_2d_impl(int do_verification,
int init_method, int init_method,
bool do_log, bool do_log,
int nrepeat, bool time_kernel,
int M, int M,
int N, int N,
int K, int K,
...@@ -259,7 +259,8 @@ void profile_gemm_bias_2d_impl(int do_verification, ...@@ -259,7 +259,8 @@ void profile_gemm_bias_2d_impl(int do_verification,
{ {
std::string gemm_name = gemm_ptr->GetTypeString(); std::string gemm_name = gemm_ptr->GetTypeString();
float ave_time = invoker_ptr->Run(argument_ptr.get(), nrepeat); float ave_time =
invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
std::size_t flop = std::size_t(2) * M * N * K; std::size_t flop = std::size_t(2) * M * N * K;
......
...@@ -48,7 +48,7 @@ template <typename ADataType, ...@@ -48,7 +48,7 @@ template <typename ADataType,
void profile_gemm_bias_relu_add_impl(int do_verification, void profile_gemm_bias_relu_add_impl(int do_verification,
int init_method, int init_method,
bool do_log, bool do_log,
int nrepeat, bool time_kernel,
int M, int M,
int N, int N,
int K, int K,
...@@ -232,7 +232,8 @@ void profile_gemm_bias_relu_add_impl(int do_verification, ...@@ -232,7 +232,8 @@ void profile_gemm_bias_relu_add_impl(int do_verification,
{ {
std::string gemm_name = gemm_ptr->GetTypeString(); std::string gemm_name = gemm_ptr->GetTypeString();
float ave_time = invoker_ptr->Run(argument_ptr.get(), nrepeat); float ave_time =
invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
std::size_t flop = std::size_t(2) * M * N * K; std::size_t flop = std::size_t(2) * M * N * K;
......
...@@ -48,7 +48,7 @@ template <typename ADataType, ...@@ -48,7 +48,7 @@ template <typename ADataType,
void profile_gemm_bias_relu_impl(int do_verification, void profile_gemm_bias_relu_impl(int do_verification,
int init_method, int init_method,
bool do_log, bool do_log,
int nrepeat, bool time_kernel,
int M, int M,
int N, int N,
int K, int K,
...@@ -212,7 +212,8 @@ void profile_gemm_bias_relu_impl(int do_verification, ...@@ -212,7 +212,8 @@ void profile_gemm_bias_relu_impl(int do_verification,
{ {
std::string gemm_name = gemm_ptr->GetTypeString(); std::string gemm_name = gemm_ptr->GetTypeString();
float ave_time = invoker_ptr->Run(argument_ptr.get(), nrepeat); float ave_time =
invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
std::size_t flop = std::size_t(2) * M * N * K; std::size_t flop = std::size_t(2) * M * N * K;
......
...@@ -91,7 +91,7 @@ template <typename ADataType, ...@@ -91,7 +91,7 @@ template <typename ADataType,
void profile_gemm_impl(int do_verification, void profile_gemm_impl(int do_verification,
int init_method, int init_method,
bool do_log, bool do_log,
int nrepeat, bool time_kernel,
int M, int M,
int N, int N,
int K, int K,
...@@ -416,7 +416,8 @@ void profile_gemm_impl(int do_verification, ...@@ -416,7 +416,8 @@ void profile_gemm_impl(int do_verification,
std::string gemm_name = gemm_ptr->GetTypeString(); std::string gemm_name = gemm_ptr->GetTypeString();
float ave_time = invoker_ptr->Run(argument_ptr.get(), nrepeat); float ave_time =
invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
std::size_t flop = std::size_t(2) * M * N * K; std::size_t flop = std::size_t(2) * M * N * K;
......
...@@ -16,11 +16,21 @@ namespace tensor_operation { ...@@ -16,11 +16,21 @@ namespace tensor_operation {
namespace device { namespace device {
namespace device_gemm_instance { namespace device_gemm_instance {
using F32 = float;
using F16 = ck::half_t;
using DPtrsGlobal = ck::Tuple<F32*, F32*>;
using Identity = ck::tensor_operation::element_wise::UnaryIdentic<F32, F32, false>;
using Square = ck::tensor_operation::element_wise::UnarySquare<F32, F32, false>;
using DInElementOps = ck::Tuple<Identity, Square>;
using DOutElementOps = ck::Tuple<Identity, Identity>;
using DeviceGemmReduceNoOpPtr = ck::tensor_operation::device::DeviceGemmReducePtr< using DeviceGemmReduceNoOpPtr = ck::tensor_operation::device::DeviceGemmReducePtr<
DPtrsGlobal,
ck::tensor_operation::element_wise::PassThrough, ck::tensor_operation::element_wise::PassThrough,
ck::tensor_operation::element_wise::PassThrough, ck::tensor_operation::element_wise::PassThrough,
ck::tensor_operation::element_wise::PassThrough, ck::tensor_operation::element_wise::PassThrough,
ck::tensor_operation::element_wise::UnarySquare<float, float, false>>; DInElementOps,
DOutElementOps>;
void add_device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_mk_kn_mn_instances( void add_device_gemm_reduce_xdl_cshuffle_f16_f16_f16_f32_f32_mk_kn_mn_instances(
std::vector<DeviceGemmReduceNoOpPtr>&); std::vector<DeviceGemmReduceNoOpPtr>&);
...@@ -52,7 +62,7 @@ template <typename ADataType, ...@@ -52,7 +62,7 @@ template <typename ADataType,
bool profile_gemm_reduce_impl(int do_verification, bool profile_gemm_reduce_impl(int do_verification,
int init_method, int init_method,
bool do_log, bool do_log,
int nrepeat, bool time_kernel,
int M, int M,
int N, int N,
int K, int K,
...@@ -112,19 +122,25 @@ bool profile_gemm_reduce_impl(int do_verification, ...@@ -112,19 +122,25 @@ bool profile_gemm_reduce_impl(int do_verification,
b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5}, num_thread); b_k_n.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5}, num_thread);
} }
using AElementOp = ck::tensor_operation::element_wise::PassThrough; using AElementOp = ck::tensor_operation::element_wise::PassThrough;
using BElementOp = ck::tensor_operation::element_wise::PassThrough; using BElementOp = ck::tensor_operation::element_wise::PassThrough;
using CElementOp = ck::tensor_operation::element_wise::PassThrough; using CElementOp = ck::tensor_operation::element_wise::PassThrough;
using D0ReduceOp = ck::reduce::Add<float>; using D0ReduceOp = ck::reduce::Add<float>;
using D1ReduceOp = ck::reduce::Add<float>; using D1ReduceOp = ck::reduce::Add<float>;
using D1ElementOp = ck::tensor_operation::element_wise::UnarySquare<float, float, false>; using UnaryIdenticElementOp =
ck::tensor_operation::element_wise::UnaryIdentic<float, float, false>;
const auto a_element_op = AElementOp{}; using UnarySquareElementOp =
const auto b_element_op = BElementOp{}; ck::tensor_operation::element_wise::UnarySquare<float, float, false>;
const auto c_element_op = CElementOp{}; using DxsInElementOps = ck::Tuple<UnaryIdenticElementOp, UnarySquareElementOp>;
const auto d0_reduce_op = D0ReduceOp{}; using DxsOutElementOps = ck::Tuple<UnaryIdenticElementOp, UnaryIdenticElementOp>;
const auto d1_reduce_op = D1ReduceOp{};
const auto d1_element_op = D1ElementOp{}; const auto a_element_op = AElementOp{};
const auto b_element_op = BElementOp{};
const auto c_element_op = CElementOp{};
const auto dxs_in_element_op = DxsInElementOps{};
const auto dxs_out_element_op = DxsOutElementOps{};
const auto d0_reduce_op = D0ReduceOp{};
const auto d1_reduce_op = D1ReduceOp{};
if(do_verification) if(do_verification)
{ {
...@@ -149,7 +165,7 @@ bool profile_gemm_reduce_impl(int do_verification, ...@@ -149,7 +165,7 @@ bool profile_gemm_reduce_impl(int do_verification,
float d0_val = ck::type_convert<float>(c_m_n_host_result(m, n)); float d0_val = ck::type_convert<float>(c_m_n_host_result(m, n));
float d1_val; float d1_val;
d1_element_op(d1_val, d0_val); UnarySquareElementOp{}(d1_val, d0_val);
d0_reduce_op(d0_acc, d0_val); d0_reduce_op(d0_acc, d0_val);
d1_reduce_op(d1_acc, d1_val); d1_reduce_op(d1_acc, d1_val);
} }
...@@ -165,6 +181,9 @@ bool profile_gemm_reduce_impl(int do_verification, ...@@ -165,6 +181,9 @@ bool profile_gemm_reduce_impl(int do_verification,
DeviceMem d0_device_buf(sizeof(DDataType) * d0_m_device_result.mDesc.GetElementSpace()); DeviceMem d0_device_buf(sizeof(DDataType) * d0_m_device_result.mDesc.GetElementSpace());
DeviceMem d1_device_buf(sizeof(DDataType) * d1_m_device_result.mDesc.GetElementSpace()); DeviceMem d1_device_buf(sizeof(DDataType) * d1_m_device_result.mDesc.GetElementSpace());
auto dxs_global = ck::make_tuple(static_cast<DDataType*>(d0_device_buf.GetDeviceBuffer()),
static_cast<DDataType*>(d1_device_buf.GetDeviceBuffer()));
a_device_buf.ToDevice(a_m_k.mData.data()); a_device_buf.ToDevice(a_m_k.mData.data());
b_device_buf.ToDevice(b_k_n.mData.data()); b_device_buf.ToDevice(b_k_n.mData.data());
...@@ -226,8 +245,7 @@ bool profile_gemm_reduce_impl(int do_verification, ...@@ -226,8 +245,7 @@ bool profile_gemm_reduce_impl(int do_verification,
gemm_ptr->MakeArgumentPointer(static_cast<ADataType*>(a_device_buf.GetDeviceBuffer()), gemm_ptr->MakeArgumentPointer(static_cast<ADataType*>(a_device_buf.GetDeviceBuffer()),
static_cast<BDataType*>(b_device_buf.GetDeviceBuffer()), static_cast<BDataType*>(b_device_buf.GetDeviceBuffer()),
static_cast<CDataType*>(c_device_buf.GetDeviceBuffer()), static_cast<CDataType*>(c_device_buf.GetDeviceBuffer()),
static_cast<DDataType*>(d0_device_buf.GetDeviceBuffer()), dxs_global,
static_cast<DDataType*>(d1_device_buf.GetDeviceBuffer()),
M, M,
N, N,
K, K,
...@@ -237,42 +255,25 @@ bool profile_gemm_reduce_impl(int do_verification, ...@@ -237,42 +255,25 @@ bool profile_gemm_reduce_impl(int do_verification,
a_element_op, a_element_op,
b_element_op, b_element_op,
c_element_op, c_element_op,
d1_element_op); dxs_in_element_op,
dxs_out_element_op);
auto invoker_ptr = gemm_ptr->MakeInvokerPointer(); auto invoker_ptr = gemm_ptr->MakeInvokerPointer();
if(gemm_ptr->IsSupportedArgument(argument_ptr.get())) if(gemm_ptr->IsSupportedArgument(argument_ptr.get()))
{ {
// warm up // init DO, D1 to 0
invoker_ptr->Run(argument_ptr.get()); d0_device_buf.SetZero();
d1_device_buf.SetZero();
// timing
float total_time = 0;
for(int i = 0; i < nrepeat; ++i)
{
// init DO, D1 to 0
d0_device_buf.SetZero();
d1_device_buf.SetZero();
KernelTimer timer;
timer.Start();
invoker_ptr->Run(argument_ptr.get());
timer.End();
total_time += timer.GetElapsedTime();
}
float ave_time = total_time / nrepeat; float ave_time =
invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
std::string gemm_name = gemm_ptr->GetTypeString(); std::string gemm_name = gemm_ptr->GetTypeString();
std::size_t flop = std::size_t(2) * M * N * K; std::size_t flop = std::size_t(2) * M * N * K;
std::size_t num_btype = sizeof(ADataType) * M * K + sizeof(BDataType) * K * M + std::size_t num_btype = sizeof(ADataType) * M * K + sizeof(BDataType) * K * N +
sizeof(CDataType) * M * N + sizeof(CDataType) * N; sizeof(CDataType) * M * N + sizeof(CDataType) * N;
float tflops = static_cast<float>(flop) / 1.E9 / ave_time; float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
......
...@@ -49,7 +49,7 @@ template <typename ADataType, ...@@ -49,7 +49,7 @@ template <typename ADataType,
void profile_grouped_gemm_impl(int do_verification, void profile_grouped_gemm_impl(int do_verification,
int init_method, int init_method,
bool do_log, bool do_log,
int nrepeat, bool time_kernel,
const std::vector<int>& Ms, const std::vector<int>& Ms,
const std::vector<int>& Ns, const std::vector<int>& Ns,
const std::vector<int>& Ks, const std::vector<int>& Ks,
...@@ -231,7 +231,8 @@ void profile_grouped_gemm_impl(int do_verification, ...@@ -231,7 +231,8 @@ void profile_grouped_gemm_impl(int do_verification,
{ {
std::string gemm_name = gemm_ptr->GetTypeString(); std::string gemm_name = gemm_ptr->GetTypeString();
float ave_time = invoker_ptr->Run(argument_ptr.get(), nrepeat); float ave_time =
invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
std::size_t flop = 0, num_btype = 0; std::size_t flop = 0, num_btype = 0;
for(std::size_t i = 0; i < gemm_shapes.size(); i++) for(std::size_t i = 0; i < gemm_shapes.size(); i++)
......
...@@ -157,7 +157,7 @@ void profile_reduce_impl_impl(bool do_verification, ...@@ -157,7 +157,7 @@ void profile_reduce_impl_impl(bool do_verification,
int init_method, int init_method,
bool do_log, bool do_log,
bool do_dumpout, bool do_dumpout,
int nrepeat, bool time_kernel,
const std::vector<size_t>& inLengths, const std::vector<size_t>& inLengths,
const std::vector<int>& reduceDims, const std::vector<int>& reduceDims,
float alpha, float alpha,
...@@ -430,7 +430,8 @@ void profile_reduce_impl_impl(bool do_verification, ...@@ -430,7 +430,8 @@ void profile_reduce_impl_impl(bool do_verification,
auto invoker_ptr = reduce_ptr->MakeInvokerPointer(); auto invoker_ptr = reduce_ptr->MakeInvokerPointer();
float avg_time = invoker_ptr->Run(argument_ptr.get(), nrepeat); float avg_time =
invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
std::size_t num_bytes = std::size_t num_bytes =
invariant_total_length * reduce_total_length * sizeof(InDataType) + invariant_total_length * reduce_total_length * sizeof(InDataType) +
...@@ -516,7 +517,8 @@ void profile_reduce_impl_impl(bool do_verification, ...@@ -516,7 +517,8 @@ void profile_reduce_impl_impl(bool do_verification,
auto invoker_ptr = reduce_ptr->MakeInvokerPointer(); auto invoker_ptr = reduce_ptr->MakeInvokerPointer();
float avg_time = invoker_ptr->Run(argument_ptr.get(), nrepeat); float avg_time =
invoker_ptr->Run(argument_ptr.get(), StreamConfig{nullptr, time_kernel});
std::size_t num_bytes = std::size_t num_bytes =
invariant_total_length * reduce_total_length * sizeof(InDataType) + invariant_total_length * reduce_total_length * sizeof(InDataType) +
...@@ -554,7 +556,8 @@ void profile_reduce_impl_impl(bool do_verification, ...@@ -554,7 +556,8 @@ void profile_reduce_impl_impl(bool do_verification,
auto invoker2_ptr = reduce2_ptr->MakeInvokerPointer(); auto invoker2_ptr = reduce2_ptr->MakeInvokerPointer();
float avg_time_2 = invoker2_ptr->Run(argument2_ptr.get(), nrepeat); float avg_time_2 =
invoker2_ptr->Run(argument2_ptr.get(), StreamConfig{nullptr, time_kernel});
std::size_t num_bytes_2 = std::size_t num_bytes_2 =
static_cast<size_t>(inLengths2[0]) * inLengths2[1] * sizeof(AccDataType); static_cast<size_t>(inLengths2[0]) * inLengths2[1] * sizeof(AccDataType);
...@@ -625,7 +628,7 @@ void profile_reduce_impl(bool do_verification, ...@@ -625,7 +628,7 @@ void profile_reduce_impl(bool do_verification,
int init_method, int init_method,
bool do_log, bool do_log,
bool do_dumpout, bool do_dumpout,
int nrepeat, bool time_kernel,
const std::vector<size_t>& inLengths, const std::vector<size_t>& inLengths,
const std::vector<int>& reduceDims, const std::vector<int>& reduceDims,
ReduceTensorOp ReduceOpId, ReduceTensorOp ReduceOpId,
...@@ -663,7 +666,7 @@ void profile_reduce_impl(bool do_verification, ...@@ -663,7 +666,7 @@ void profile_reduce_impl(bool do_verification,
init_method, init_method,
do_log, do_log,
do_dumpout, do_dumpout,
nrepeat, time_kernel,
inLengths, inLengths,
reduceDims, reduceDims,
alpha, alpha,
......
...@@ -48,8 +48,8 @@ int profile_batched_gemm(int argc, char* argv[]) ...@@ -48,8 +48,8 @@ int profile_batched_gemm(int argc, char* argv[])
printf(" 3: A[g, k, m] * B[g, n, k] = C[g, m, n])\n"); printf(" 3: A[g, k, m] * B[g, n, k] = C[g, m, n])\n");
printf("arg4: verification (0: no; 1: yes)\n"); printf("arg4: verification (0: no; 1: yes)\n");
printf("arg5: initialization (0: no init; 1: integer value; 2: decimal value)\n"); printf("arg5: initialization (0: no init; 1: integer value; 2: decimal value)\n");
printf("arg8: print tensor value (0: no; 1: yes)\n"); printf("arg6: print tensor value (0: no; 1: yes)\n");
printf("arg7: run kernel # of times (>1)\n"); printf("arg7: time kernel (0=n0, 1=yes)\n");
printf("arg8 to 14: M, N, K, StrideA, StrideB, StrideC, BatchCount\n"); printf("arg8 to 14: M, N, K, StrideA, StrideB, StrideC, BatchCount\n");
exit(1); exit(1);
} }
...@@ -59,7 +59,7 @@ int profile_batched_gemm(int argc, char* argv[]) ...@@ -59,7 +59,7 @@ int profile_batched_gemm(int argc, char* argv[])
const bool do_verification = std::stoi(argv[4]); const bool do_verification = std::stoi(argv[4]);
const int init_method = std::stoi(argv[5]); const int init_method = std::stoi(argv[5]);
const bool do_log = std::stoi(argv[6]); const bool do_log = std::stoi(argv[6]);
const int nrepeat = std::stoi(argv[7]); const bool time_kernel = std::stoi(argv[7]);
const int M = std::stoi(argv[8]); const int M = std::stoi(argv[8]);
const int N = std::stoi(argv[9]); const int N = std::stoi(argv[9]);
...@@ -82,7 +82,7 @@ int profile_batched_gemm(int argc, char* argv[]) ...@@ -82,7 +82,7 @@ int profile_batched_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -102,7 +102,7 @@ int profile_batched_gemm(int argc, char* argv[]) ...@@ -102,7 +102,7 @@ int profile_batched_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -122,7 +122,7 @@ int profile_batched_gemm(int argc, char* argv[]) ...@@ -122,7 +122,7 @@ int profile_batched_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -142,7 +142,7 @@ int profile_batched_gemm(int argc, char* argv[]) ...@@ -142,7 +142,7 @@ int profile_batched_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -162,7 +162,7 @@ int profile_batched_gemm(int argc, char* argv[]) ...@@ -162,7 +162,7 @@ int profile_batched_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -182,7 +182,7 @@ int profile_batched_gemm(int argc, char* argv[]) ...@@ -182,7 +182,7 @@ int profile_batched_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -202,7 +202,7 @@ int profile_batched_gemm(int argc, char* argv[]) ...@@ -202,7 +202,7 @@ int profile_batched_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -222,7 +222,7 @@ int profile_batched_gemm(int argc, char* argv[]) ...@@ -222,7 +222,7 @@ int profile_batched_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -242,7 +242,7 @@ int profile_batched_gemm(int argc, char* argv[]) ...@@ -242,7 +242,7 @@ int profile_batched_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -262,7 +262,7 @@ int profile_batched_gemm(int argc, char* argv[]) ...@@ -262,7 +262,7 @@ int profile_batched_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -282,7 +282,7 @@ int profile_batched_gemm(int argc, char* argv[]) ...@@ -282,7 +282,7 @@ int profile_batched_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -302,7 +302,7 @@ int profile_batched_gemm(int argc, char* argv[]) ...@@ -302,7 +302,7 @@ int profile_batched_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -322,7 +322,7 @@ int profile_batched_gemm(int argc, char* argv[]) ...@@ -322,7 +322,7 @@ int profile_batched_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -342,7 +342,7 @@ int profile_batched_gemm(int argc, char* argv[]) ...@@ -342,7 +342,7 @@ int profile_batched_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -362,7 +362,7 @@ int profile_batched_gemm(int argc, char* argv[]) ...@@ -362,7 +362,7 @@ int profile_batched_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -382,7 +382,7 @@ int profile_batched_gemm(int argc, char* argv[]) ...@@ -382,7 +382,7 @@ int profile_batched_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
......
...@@ -33,8 +33,8 @@ int profile_batched_gemm_reduce(int argc, char* argv[]) ...@@ -33,8 +33,8 @@ int profile_batched_gemm_reduce(int argc, char* argv[])
printf(" 3: A[k, m] * B[n, k] = C[m, n])\n"); printf(" 3: A[k, m] * B[n, k] = C[m, n])\n");
printf("arg4: verification (0: no; 1: yes)\n"); printf("arg4: verification (0: no; 1: yes)\n");
printf("arg5: initialization (0: no init; 1: integer value; 2: decimal value)\n"); printf("arg5: initialization (0: no init; 1: integer value; 2: decimal value)\n");
printf("arg8: print tensor value (0: no; 1: yes)\n"); printf("arg6: print tensor value (0: no; 1: yes)\n");
printf("arg7: run kernel # of times (>1)\n"); printf("arg7: time kernel (0=n0, 1=yes)\n");
printf("arg8 to 14: M, N, K, StrideA, StrideB, StrideC, BatchCount\n"); printf("arg8 to 14: M, N, K, StrideA, StrideB, StrideC, BatchCount\n");
printf("arg15: split k into mulitiple batch\n"); printf("arg15: split k into mulitiple batch\n");
exit(1); exit(1);
...@@ -45,7 +45,7 @@ int profile_batched_gemm_reduce(int argc, char* argv[]) ...@@ -45,7 +45,7 @@ int profile_batched_gemm_reduce(int argc, char* argv[])
const bool do_verification = std::stoi(argv[4]); const bool do_verification = std::stoi(argv[4]);
const int init_method = std::stoi(argv[5]); const int init_method = std::stoi(argv[5]);
const bool do_log = std::stoi(argv[6]); const bool do_log = std::stoi(argv[6]);
const int nrepeat = std::stoi(argv[7]); const bool time_kernel = std::stoi(argv[7]);
const int M = std::stoi(argv[8]); const int M = std::stoi(argv[8]);
const int N = std::stoi(argv[9]); const int N = std::stoi(argv[9]);
...@@ -69,7 +69,7 @@ int profile_batched_gemm_reduce(int argc, char* argv[]) ...@@ -69,7 +69,7 @@ int profile_batched_gemm_reduce(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -91,7 +91,7 @@ int profile_batched_gemm_reduce(int argc, char* argv[]) ...@@ -91,7 +91,7 @@ int profile_batched_gemm_reduce(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -113,7 +113,7 @@ int profile_batched_gemm_reduce(int argc, char* argv[]) ...@@ -113,7 +113,7 @@ int profile_batched_gemm_reduce(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -135,7 +135,7 @@ int profile_batched_gemm_reduce(int argc, char* argv[]) ...@@ -135,7 +135,7 @@ int profile_batched_gemm_reduce(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
......
#include <iostream>
#include <numeric>
#include <initializer_list>
#include <cstdlib>
#include <stdlib.h>
#include <half.hpp>
#include "profile_conv_bwd_data_impl.hpp"
enum struct ConvDataType
{
F32_F32_F32, // 0
F16_F16_F16, // 1
BF16_BF16_BF16, // 2
INT8_INT8_INT8, // 3
};
enum struct ConvInputLayout
{
NCHW, // 0
NHWC, // 1
};
enum struct ConvWeightLayout
{
KCYX, // 0
KYXC, // 1
};
enum struct ConvOutputLayout
{
NKHW, // 0
NHWK, // 1
};
int profile_conv_bwd_data(int argc, char* argv[])
{
if(argc != 25)
{
printf("arg1: tensor operation (conv_bwd: BackwardConvolution)\n");
printf("arg2: data type (0: fp32; 1: fp16)\n");
printf("arg3: input tensor layout (0: NCHW; 1: NHWC)\n");
printf("arg4: weight tensor layout (0: KCYX; 1: KYXC)\n");
printf("arg5: output tensor layout (0: NKHW; 1: NHWK)\n");
printf("arg6: verification (0: no; 1: yes)\n");
printf("arg7: initialization (0: no init; 1: integer value; 2: decimal value)\n");
printf("arg8: print tensor value (0: no; 1: yes)\n");
printf("arg9: run kernel # of times (>1)\n");
printf("arg10 to 24: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, "
"RightPx\n");
exit(1);
}
const auto data_type = static_cast<ConvDataType>(std::stoi(argv[2]));
const auto in_layout = static_cast<ConvInputLayout>(std::stoi(argv[3]));
const auto wei_layout = static_cast<ConvWeightLayout>(std::stoi(argv[4]));
const auto out_layout = static_cast<ConvOutputLayout>(std::stoi(argv[5]));
const bool do_verification = std::stoi(argv[6]);
const int init_method = std::stoi(argv[7]);
const bool do_log = std::stoi(argv[8]);
const int nrepeat = std::stoi(argv[9]);
const ck::index_t N = std::stoi(argv[10]);
const ck::index_t K = std::stoi(argv[11]);
const ck::index_t C = std::stoi(argv[12]);
const ck::index_t Y = std::stoi(argv[13]);
const ck::index_t X = std::stoi(argv[14]);
const ck::index_t Hi = std::stoi(argv[15]);
const ck::index_t Wi = std::stoi(argv[16]);
const ck::index_t conv_stride_h = std::stoi(argv[17]);
const ck::index_t conv_stride_w = std::stoi(argv[18]);
const ck::index_t conv_dilation_h = std::stoi(argv[19]);
const ck::index_t conv_dilation_w = std::stoi(argv[20]);
const ck::index_t in_left_pad_h = std::stoi(argv[21]);
const ck::index_t in_left_pad_w = std::stoi(argv[22]);
const ck::index_t in_right_pad_h = std::stoi(argv[23]);
const ck::index_t in_right_pad_w = std::stoi(argv[24]);
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;
if(data_type == ConvDataType::F32_F32_F32 && in_layout == ConvInputLayout::NHWC &&
wei_layout == ConvWeightLayout::KYXC && out_layout == ConvOutputLayout::NHWK)
{
ck::profiler::profile_conv_bwd_data_impl<2,
float,
float,
float,
float,
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::F16_F16_F16 && in_layout == ConvInputLayout::NHWC &&
wei_layout == ConvWeightLayout::KYXC && out_layout == ConvOutputLayout::NHWK)
{
ck::profiler::profile_conv_bwd_data_impl<2,
ck::half_t,
ck::half_t,
ck::half_t,
float,
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::BF16_BF16_BF16 && in_layout == ConvInputLayout::NHWC &&
wei_layout == ConvWeightLayout::KYXC && out_layout == ConvOutputLayout::NHWK)
{
ck::profiler::profile_conv_bwd_data_impl<2,
uint16_t,
uint16_t,
uint16_t,
float,
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_bwd_data_impl<2,
int8_t,
int8_t,
int8_t,
int32_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");
}
return 1;
}
...@@ -58,7 +58,7 @@ int profile_conv_bwd_weight(int argc, char* argv[]) ...@@ -58,7 +58,7 @@ int profile_conv_bwd_weight(int argc, char* argv[])
const bool do_verification = std::stoi(argv[6]); const bool do_verification = std::stoi(argv[6]);
const int init_method = std::stoi(argv[7]); const int init_method = std::stoi(argv[7]);
const bool do_log = std::stoi(argv[8]); const bool do_log = std::stoi(argv[8]);
const int nrepeat = std::stoi(argv[9]); const bool time_kernel = std::stoi(argv[9]);
const ck::index_t N = std::stoi(argv[10]); const ck::index_t N = std::stoi(argv[10]);
const ck::index_t K = std::stoi(argv[11]); const ck::index_t K = std::stoi(argv[11]);
...@@ -98,7 +98,7 @@ int profile_conv_bwd_weight(int argc, char* argv[]) ...@@ -98,7 +98,7 @@ int profile_conv_bwd_weight(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
N, N,
K, K,
C, C,
...@@ -124,7 +124,7 @@ int profile_conv_bwd_weight(int argc, char* argv[]) ...@@ -124,7 +124,7 @@ int profile_conv_bwd_weight(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
N, N,
K, K,
C, C,
......
...@@ -42,7 +42,7 @@ int profile_conv_fwd_bias_relu(int argc, char* argv[]) ...@@ -42,7 +42,7 @@ int profile_conv_fwd_bias_relu(int argc, char* argv[])
printf("arg6: verification (0: no; 1: yes)\n"); printf("arg6: verification (0: no; 1: yes)\n");
printf("arg7: initialization (0: no init; 1: integer value; 2: decimal value)\n"); printf("arg7: initialization (0: no init; 1: integer value; 2: decimal value)\n");
printf("arg8: print tensor value (0: no; 1: yes)\n"); printf("arg8: print tensor value (0: no; 1: yes)\n");
printf("arg9: run kernel # of times (>1)\n"); printf("arg9: time kernel (0=n0, 1=yes)\n");
printf("arg10 to 24: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, " printf("arg10 to 24: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, "
"RightPx\n"); "RightPx\n");
exit(1); exit(1);
...@@ -55,7 +55,7 @@ int profile_conv_fwd_bias_relu(int argc, char* argv[]) ...@@ -55,7 +55,7 @@ int profile_conv_fwd_bias_relu(int argc, char* argv[])
const bool do_verification = std::stoi(argv[6]); const bool do_verification = std::stoi(argv[6]);
const int init_method = std::stoi(argv[7]); const int init_method = std::stoi(argv[7]);
const bool do_log = std::stoi(argv[8]); const bool do_log = std::stoi(argv[8]);
const int nrepeat = std::stoi(argv[9]); const bool time_kernel = std::stoi(argv[9]);
const ck::index_t N = std::stoi(argv[10]); const ck::index_t N = std::stoi(argv[10]);
const ck::index_t K = std::stoi(argv[11]); const ck::index_t K = std::stoi(argv[11]);
...@@ -93,7 +93,7 @@ int profile_conv_fwd_bias_relu(int argc, char* argv[]) ...@@ -93,7 +93,7 @@ int profile_conv_fwd_bias_relu(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
N, N,
K, K,
C, C,
......
...@@ -43,7 +43,7 @@ int profile_conv_fwd_bias_relu_add(int argc, char* argv[]) ...@@ -43,7 +43,7 @@ int profile_conv_fwd_bias_relu_add(int argc, char* argv[])
printf("arg6: verification (0: no; 1: yes)\n"); printf("arg6: verification (0: no; 1: yes)\n");
printf("arg7: initialization (0: no init; 1: integer value; 2: decimal value)\n"); printf("arg7: initialization (0: no init; 1: integer value; 2: decimal value)\n");
printf("arg8: print tensor value (0: no; 1: yes)\n"); printf("arg8: print tensor value (0: no; 1: yes)\n");
printf("arg9: run kernel # of times (>1)\n"); printf("arg9: time kernel (0=n0, 1=yes)\n");
printf("arg10 to 24: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, " printf("arg10 to 24: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, "
"RightPx\n"); "RightPx\n");
exit(1); exit(1);
...@@ -56,7 +56,7 @@ int profile_conv_fwd_bias_relu_add(int argc, char* argv[]) ...@@ -56,7 +56,7 @@ int profile_conv_fwd_bias_relu_add(int argc, char* argv[])
const bool do_verification = std::stoi(argv[6]); const bool do_verification = std::stoi(argv[6]);
const int init_method = std::stoi(argv[7]); const int init_method = std::stoi(argv[7]);
const bool do_log = std::stoi(argv[8]); const bool do_log = std::stoi(argv[8]);
const int nrepeat = std::stoi(argv[9]); const bool time_kernel = std::stoi(argv[9]);
const ck::index_t N = std::stoi(argv[10]); const ck::index_t N = std::stoi(argv[10]);
const ck::index_t K = std::stoi(argv[11]); const ck::index_t K = std::stoi(argv[11]);
...@@ -94,7 +94,7 @@ int profile_conv_fwd_bias_relu_add(int argc, char* argv[]) ...@@ -94,7 +94,7 @@ int profile_conv_fwd_bias_relu_add(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
N, N,
K, K,
C, C,
......
...@@ -43,7 +43,7 @@ int profile_conv_fwd_bias_relu_atomic_add(int argc, char* argv[]) ...@@ -43,7 +43,7 @@ int profile_conv_fwd_bias_relu_atomic_add(int argc, char* argv[])
printf("arg6: verification (0: no; 1: yes)\n"); printf("arg6: verification (0: no; 1: yes)\n");
printf("arg7: initialization (0: no init; 1: integer value; 2: decimal value)\n"); printf("arg7: initialization (0: no init; 1: integer value; 2: decimal value)\n");
printf("arg8: print tensor value (0: no; 1: yes)\n"); printf("arg8: print tensor value (0: no; 1: yes)\n");
printf("arg9: run kernel # of times (>1)\n"); printf("arg9: time kernel (0=n0, 1=yes)\n");
printf("arg10 to 24: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, " printf("arg10 to 24: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, "
"RightPx\n"); "RightPx\n");
exit(1); exit(1);
...@@ -56,7 +56,7 @@ int profile_conv_fwd_bias_relu_atomic_add(int argc, char* argv[]) ...@@ -56,7 +56,7 @@ int profile_conv_fwd_bias_relu_atomic_add(int argc, char* argv[])
const bool do_verification = std::stoi(argv[6]); const bool do_verification = std::stoi(argv[6]);
const int init_method = std::stoi(argv[7]); const int init_method = std::stoi(argv[7]);
const bool do_log = std::stoi(argv[8]); const bool do_log = std::stoi(argv[8]);
const int nrepeat = std::stoi(argv[9]); const bool time_kernel = std::stoi(argv[9]);
const ck::index_t N = std::stoi(argv[10]); const ck::index_t N = std::stoi(argv[10]);
const ck::index_t K = std::stoi(argv[11]); const ck::index_t K = std::stoi(argv[11]);
...@@ -95,7 +95,7 @@ int profile_conv_fwd_bias_relu_atomic_add(int argc, char* argv[]) ...@@ -95,7 +95,7 @@ int profile_conv_fwd_bias_relu_atomic_add(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
N, N,
K, K,
C, C,
......
...@@ -95,7 +95,7 @@ int profile_convnd_bwd_data(int argc, char* argv[], int num_dim_spatial) ...@@ -95,7 +95,7 @@ int profile_convnd_bwd_data(int argc, char* argv[], int num_dim_spatial)
printf("arg6: verification (0: no; 1: yes)\n"); printf("arg6: verification (0: no; 1: yes)\n");
printf("arg7: initialization (0: no init; 1: integer value; 2: decimal value)\n"); printf("arg7: initialization (0: no init; 1: integer value; 2: decimal value)\n");
printf("arg8: print tensor value (0: no; 1: yes)\n"); printf("arg8: print tensor value (0: no; 1: yes)\n");
printf("arg9: run kernel # of times (>1)\n"); printf("arg9: time kernel (0=n0, 1=yes)\n");
printf("arg10 to 24: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, " printf("arg10 to 24: N, K, C, Y, X, Hi, Wi, Sy, Sx, Dy, Dx, LeftPy, LeftPx, RightPy, "
"RightPx\n"); "RightPx\n");
return 1; return 1;
...@@ -108,7 +108,7 @@ int profile_convnd_bwd_data(int argc, char* argv[], int num_dim_spatial) ...@@ -108,7 +108,7 @@ int profile_convnd_bwd_data(int argc, char* argv[], int num_dim_spatial)
const bool do_verification = std::stoi(argv[6]); const bool do_verification = std::stoi(argv[6]);
const int init_method = std::stoi(argv[7]); const int init_method = std::stoi(argv[7]);
const bool do_log = std::stoi(argv[8]); const bool do_log = std::stoi(argv[8]);
const int nrepeat = std::stoi(argv[9]); const bool time_kernel = std::stoi(argv[9]);
ck::utils::conv::ConvParams params = parse_conv_params(num_dim_spatial, argv, preParams); ck::utils::conv::ConvParams params = parse_conv_params(num_dim_spatial, argv, preParams);
...@@ -132,7 +132,7 @@ int profile_convnd_bwd_data(int argc, char* argv[], int num_dim_spatial) ...@@ -132,7 +132,7 @@ int profile_convnd_bwd_data(int argc, char* argv[], int num_dim_spatial)
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
params.N_, params.N_,
params.K_, params.K_,
params.C_, params.C_,
...@@ -157,7 +157,7 @@ int profile_convnd_bwd_data(int argc, char* argv[], int num_dim_spatial) ...@@ -157,7 +157,7 @@ int profile_convnd_bwd_data(int argc, char* argv[], int num_dim_spatial)
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
params.N_, params.N_,
params.K_, params.K_,
params.C_, params.C_,
...@@ -182,7 +182,7 @@ int profile_convnd_bwd_data(int argc, char* argv[], int num_dim_spatial) ...@@ -182,7 +182,7 @@ int profile_convnd_bwd_data(int argc, char* argv[], int num_dim_spatial)
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
params.N_, params.N_,
params.K_, params.K_,
params.C_, params.C_,
......
...@@ -119,7 +119,7 @@ template <int NDim, ...@@ -119,7 +119,7 @@ template <int NDim,
void profile_convnd_instances_impl(const ck::utils::conv::ConvParams& params, void profile_convnd_instances_impl(const ck::utils::conv::ConvParams& params,
bool do_verification, bool do_verification,
bool do_log, bool do_log,
int nrepeat, bool time_kernel,
int init_method, int init_method,
ConvLayouts) ConvLayouts)
{ {
...@@ -185,7 +185,7 @@ void profile_convnd_instances_impl(const ck::utils::conv::ConvParams& params, ...@@ -185,7 +185,7 @@ void profile_convnd_instances_impl(const ck::utils::conv::ConvParams& params,
reference_conv_fwd_fun); reference_conv_fwd_fun);
auto best_conf = run_engine.Profile( auto best_conf = run_engine.Profile(
conv::ConvolutionFwdInstances<InDataType, WeiDataType, OutDataType>::template Get<NDim>(), conv::ConvolutionFwdInstances<InDataType, WeiDataType, OutDataType>::template Get<NDim>(),
nrepeat, time_kernel,
do_verification, do_verification,
do_log); do_log);
...@@ -201,7 +201,7 @@ void profile_convnd_instances(ConvDataType data_type, ...@@ -201,7 +201,7 @@ void profile_convnd_instances(ConvDataType data_type,
const ck::utils::conv::ConvParams& params, const ck::utils::conv::ConvParams& params,
bool do_verification, bool do_verification,
bool do_log, bool do_log,
int nrepeat, bool time_kernel,
int init_method) int init_method)
{ {
switch(data_layout) switch(data_layout)
...@@ -214,7 +214,7 @@ void profile_convnd_instances(ConvDataType data_type, ...@@ -214,7 +214,7 @@ void profile_convnd_instances(ConvDataType data_type,
params, params,
do_verification, do_verification,
do_log, do_log,
nrepeat, time_kernel,
init_method, init_method,
ConvolutionLayouts<NDim, ConvDataLayout::NHWC>{}); ConvolutionLayouts<NDim, ConvDataLayout::NHWC>{});
break; break;
...@@ -223,7 +223,7 @@ void profile_convnd_instances(ConvDataType data_type, ...@@ -223,7 +223,7 @@ void profile_convnd_instances(ConvDataType data_type,
params, params,
do_verification, do_verification,
do_log, do_log,
nrepeat, time_kernel,
init_method, init_method,
ConvolutionLayouts<NDim, ConvDataLayout::NHWC>{}); ConvolutionLayouts<NDim, ConvDataLayout::NHWC>{});
break; break;
...@@ -232,7 +232,7 @@ void profile_convnd_instances(ConvDataType data_type, ...@@ -232,7 +232,7 @@ void profile_convnd_instances(ConvDataType data_type,
params, params,
do_verification, do_verification,
do_log, do_log,
nrepeat, time_kernel,
init_method, init_method,
ConvolutionLayouts<NDim, ConvDataLayout::NHWC>{}); ConvolutionLayouts<NDim, ConvDataLayout::NHWC>{});
break; break;
...@@ -241,7 +241,7 @@ void profile_convnd_instances(ConvDataType data_type, ...@@ -241,7 +241,7 @@ void profile_convnd_instances(ConvDataType data_type,
params, params,
do_verification, do_verification,
do_log, do_log,
nrepeat, time_kernel,
init_method, init_method,
ConvolutionLayouts<NDim, ConvDataLayout::NHWC>{}); ConvolutionLayouts<NDim, ConvDataLayout::NHWC>{});
break; break;
...@@ -256,7 +256,7 @@ void profile_convnd_instances(ConvDataType data_type, ...@@ -256,7 +256,7 @@ void profile_convnd_instances(ConvDataType data_type,
params, params,
do_verification, do_verification,
do_log, do_log,
nrepeat, time_kernel,
init_method, init_method,
ConvolutionLayouts<NDim, ConvDataLayout::NCHW>{}); ConvolutionLayouts<NDim, ConvDataLayout::NCHW>{});
break; break;
...@@ -265,7 +265,7 @@ void profile_convnd_instances(ConvDataType data_type, ...@@ -265,7 +265,7 @@ void profile_convnd_instances(ConvDataType data_type,
params, params,
do_verification, do_verification,
do_log, do_log,
nrepeat, time_kernel,
init_method, init_method,
ConvolutionLayouts<NDim, ConvDataLayout::NCHW>{}); ConvolutionLayouts<NDim, ConvDataLayout::NCHW>{});
break; break;
...@@ -274,7 +274,7 @@ void profile_convnd_instances(ConvDataType data_type, ...@@ -274,7 +274,7 @@ void profile_convnd_instances(ConvDataType data_type,
params, params,
do_verification, do_verification,
do_log, do_log,
nrepeat, time_kernel,
init_method, init_method,
ConvolutionLayouts<NDim, ConvDataLayout::NCHW>{}); ConvolutionLayouts<NDim, ConvDataLayout::NCHW>{});
break; break;
...@@ -283,7 +283,7 @@ void profile_convnd_instances(ConvDataType data_type, ...@@ -283,7 +283,7 @@ void profile_convnd_instances(ConvDataType data_type,
params, params,
do_verification, do_verification,
do_log, do_log,
nrepeat, time_kernel,
init_method, init_method,
ConvolutionLayouts<NDim, ConvDataLayout::NCHW>{}); ConvolutionLayouts<NDim, ConvDataLayout::NCHW>{});
break; break;
...@@ -304,7 +304,7 @@ int ck::profiler::profile_convnd_fwd(int argc, char* argv[]) ...@@ -304,7 +304,7 @@ int ck::profiler::profile_convnd_fwd(int argc, char* argv[])
bool do_verification{true}; bool do_verification{true};
int init_method{2}; int init_method{2};
bool do_log{false}; bool do_log{false};
int nrepeat{100}; bool time_kernel{false};
int num_dim_spatial{2}; int num_dim_spatial{2};
ConvParams params; ConvParams params;
...@@ -318,7 +318,7 @@ int ck::profiler::profile_convnd_fwd(int argc, char* argv[]) ...@@ -318,7 +318,7 @@ int ck::profiler::profile_convnd_fwd(int argc, char* argv[])
do_verification = std::stoi(argv[4]); do_verification = std::stoi(argv[4]);
init_method = std::stoi(argv[5]); init_method = std::stoi(argv[5]);
do_log = std::stoi(argv[6]); do_log = std::stoi(argv[6]);
nrepeat = std::stoi(argv[7]); time_kernel = std::stoi(argv[7]);
num_dim_spatial = std::stoi(argv[8]); num_dim_spatial = std::stoi(argv[8]);
} }
if(argc >= 10) if(argc >= 10)
...@@ -332,15 +332,15 @@ int ck::profiler::profile_convnd_fwd(int argc, char* argv[]) ...@@ -332,15 +332,15 @@ int ck::profiler::profile_convnd_fwd(int argc, char* argv[])
{ {
case 1: case 1:
profile_convnd_instances<1>( profile_convnd_instances<1>(
data_type, data_layout, params, do_verification, do_log, nrepeat, init_method); data_type, data_layout, params, do_verification, do_log, time_kernel, init_method);
break; break;
case 2: case 2:
profile_convnd_instances<2>( profile_convnd_instances<2>(
data_type, data_layout, params, do_verification, do_log, nrepeat, init_method); data_type, data_layout, params, do_verification, do_log, time_kernel, init_method);
break; break;
case 3: case 3:
profile_convnd_instances<3>( profile_convnd_instances<3>(
data_type, data_layout, params, do_verification, do_log, nrepeat, init_method); data_type, data_layout, params, do_verification, do_log, time_kernel, init_method);
break; break;
default: default:
throw std::runtime_error("profile_conv_fwd: unsupported num_dim_spatial value: " + throw std::runtime_error("profile_conv_fwd: unsupported num_dim_spatial value: " +
......
...@@ -38,8 +38,8 @@ int profile_gemm(int argc, char* argv[]) ...@@ -38,8 +38,8 @@ int profile_gemm(int argc, char* argv[])
printf(" 3: A[k, m] * B[n, k] = C[m, n])\n"); printf(" 3: A[k, m] * B[n, k] = C[m, n])\n");
printf("arg4: verification (0: no; 1: yes)\n"); printf("arg4: verification (0: no; 1: yes)\n");
printf("arg5: initialization (0: no init; 1: integer value; 2: decimal value)\n"); printf("arg5: initialization (0: no init; 1: integer value; 2: decimal value)\n");
printf("arg8: print tensor value (0: no; 1: yes)\n"); printf("arg6: print tensor value (0: no; 1: yes)\n");
printf("arg7: run kernel # of times (>1)\n"); printf("arg7: time kernel (0=n0, 1=yes)\n");
printf("arg8 to 13: M, N, K, StrideA, StrideB, StrideC\n"); printf("arg8 to 13: M, N, K, StrideA, StrideB, StrideC\n");
printf("arg14: split k into mulitiple batch\n"); printf("arg14: split k into mulitiple batch\n");
exit(1); exit(1);
...@@ -50,7 +50,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -50,7 +50,7 @@ int profile_gemm(int argc, char* argv[])
const bool do_verification = std::stoi(argv[4]); const bool do_verification = std::stoi(argv[4]);
const int init_method = std::stoi(argv[5]); const int init_method = std::stoi(argv[5]);
const bool do_log = std::stoi(argv[6]); const bool do_log = std::stoi(argv[6]);
const int nrepeat = std::stoi(argv[7]); const bool time_kernel = std::stoi(argv[7]);
const int M = std::stoi(argv[8]); const int M = std::stoi(argv[8]);
const int N = std::stoi(argv[9]); const int N = std::stoi(argv[9]);
...@@ -74,7 +74,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -74,7 +74,7 @@ int profile_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -94,7 +94,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -94,7 +94,7 @@ int profile_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -114,7 +114,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -114,7 +114,7 @@ int profile_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -134,7 +134,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -134,7 +134,7 @@ int profile_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -154,7 +154,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -154,7 +154,7 @@ int profile_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -174,7 +174,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -174,7 +174,7 @@ int profile_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -194,7 +194,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -194,7 +194,7 @@ int profile_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -214,7 +214,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -214,7 +214,7 @@ int profile_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -234,7 +234,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -234,7 +234,7 @@ int profile_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -254,7 +254,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -254,7 +254,7 @@ int profile_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -274,7 +274,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -274,7 +274,7 @@ int profile_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -294,7 +294,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -294,7 +294,7 @@ int profile_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -314,7 +314,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -314,7 +314,7 @@ int profile_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -334,7 +334,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -334,7 +334,7 @@ int profile_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -354,7 +354,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -354,7 +354,7 @@ int profile_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
...@@ -374,7 +374,7 @@ int profile_gemm(int argc, char* argv[]) ...@@ -374,7 +374,7 @@ int profile_gemm(int argc, char* argv[])
do_verification, do_verification,
init_method, init_method,
do_log, do_log,
nrepeat, time_kernel,
M, M,
N, N,
K, K,
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment