Skip to content

GitLab

Commit 1e5e2dc3 authored by muozturk's avatar muozturk
Browse files

merge

parents 571e8728 81eece66
Show whitespace changes
Inline Side-by-side
Showing with 414 additions and 337 deletions
example/64_complex_contraction_bilinear/run_complex_contraction_bilinear_example.inc 100755 → 100644
View file @ 1e5e2dc3
......@@ -154,17 +154,20 @@ int run_complex_contraction_bilinear_example(int argc, char* argv[])
DeviceMem a_device_buf_re(sizeof(ADataType) * a_ms_ks_re.mDesc.GetElementSpaceSize());
DeviceMem b_device_buf_re(sizeof(BDataType) * b_ns_ks_re.mDesc.GetElementSpaceSize());
DeviceMem d_device_buf_re(sizeof(DDataType) * d_ms_ns_re.mDesc.GetElementSpaceSize());
DeviceMem e_device_buf_re(sizeof(EDataType) * e_ms_ns_device_result_re.mDesc.GetElementSpaceSize());
DeviceMem e_device_buf_re(sizeof(EDataType) *
e_ms_ns_device_result_re.mDesc.GetElementSpaceSize());
DeviceMem a_device_buf_img(sizeof(ADataType) * a_ms_ks_img.mDesc.GetElementSpaceSize());
DeviceMem b_device_buf_img(sizeof(BDataType) * b_ns_ks_img.mDesc.GetElementSpaceSize());
DeviceMem d_device_buf_img(sizeof(DDataType) * d_ms_ns_img.mDesc.GetElementSpaceSize());
DeviceMem e_device_buf_img(sizeof(EDataType) * e_ms_ns_device_result_img.mDesc.GetElementSpaceSize());
DeviceMem e_device_buf_img(sizeof(EDataType) *
e_ms_ns_device_result_img.mDesc.GetElementSpaceSize());
// Intermediate Value For E Real and Img
DeviceMem e_device_buf_re1(sizeof(EDataType) * e_ms_ns_device_result_re.mDesc.GetElementSpaceSize());
DeviceMem e_device_buf_img1(sizeof(EDataType) * e_ms_ns_device_result_img.mDesc.GetElementSpaceSize());
DeviceMem e_device_buf_re1(sizeof(EDataType) *
e_ms_ns_device_result_re.mDesc.GetElementSpaceSize());
DeviceMem e_device_buf_img1(sizeof(EDataType) *
e_ms_ns_device_result_img.mDesc.GetElementSpaceSize());
a_device_buf_re.ToDevice(a_ms_ks_re.mData.data());
b_device_buf_re.ToDevice(b_ns_ks_re.mData.data());
......@@ -191,7 +194,8 @@ int run_complex_contraction_bilinear_example(int argc, char* argv[])
auto op = DeviceOpInstance{};
auto invoker = op.MakeInvoker();
auto argument_re1 = op.MakeArgument(a_device_buf_re.GetDeviceBuffer(),
auto argument_re1 =
op.MakeArgument(a_device_buf_re.GetDeviceBuffer(),
b_device_buf_re.GetDeviceBuffer(),
std::array<const void*, 1>{d_device_buf_re.GetDeviceBuffer()},
e_device_buf_re1.GetDeviceBuffer(),
......@@ -216,7 +220,6 @@ int run_complex_contraction_bilinear_example(int argc, char* argv[])
float ave_time_re1 = invoker.Run(argument_re1, StreamConfig{nullptr, time_kernel});
alpha = -1.f;
beta = 1.f;
......@@ -228,7 +231,8 @@ int run_complex_contraction_bilinear_example(int argc, char* argv[])
// For real Intermediate Value re_2
// auto op = DeviceOpInstance{};
// auto invoker = op.MakeInvoker();
auto argument_re2 = op.MakeArgument(a_device_buf_img.GetDeviceBuffer(),
auto argument_re2 =
op.MakeArgument(a_device_buf_img.GetDeviceBuffer(),
b_device_buf_img.GetDeviceBuffer(),
std::array<const void*, 1>{e_device_buf_re1.GetDeviceBuffer()},
e_device_buf_re.GetDeviceBuffer(),
......@@ -253,7 +257,6 @@ int run_complex_contraction_bilinear_example(int argc, char* argv[])
float ave_time_re2 = invoker.Run(argument_re2, StreamConfig{nullptr, time_kernel});
alpha = 1.f;
beta = 1.f;
......@@ -261,7 +264,8 @@ int run_complex_contraction_bilinear_example(int argc, char* argv[])
b_element_op = BElementOp{};
cde_element_op = CDEElementOp{alpha, beta};
auto argument_img1 = op.MakeArgument(a_device_buf_re.GetDeviceBuffer(),
auto argument_img1 =
op.MakeArgument(a_device_buf_re.GetDeviceBuffer(),
b_device_buf_img.GetDeviceBuffer(),
std::array<const void*, 1>{d_device_buf_img.GetDeviceBuffer()},
e_device_buf_img1.GetDeviceBuffer(),
......@@ -277,7 +281,6 @@ int run_complex_contraction_bilinear_example(int argc, char* argv[])
b_element_op,
cde_element_op);
if(!op.IsSupportedArgument(argument_img1))
{
std::cout << op.GetTypeString() << " does not support this problem" << std::endl;
......@@ -290,7 +293,8 @@ int run_complex_contraction_bilinear_example(int argc, char* argv[])
alpha = 1.f;
beta = 1.f;
auto argument_img2 = op.MakeArgument(a_device_buf_img.GetDeviceBuffer(),
auto argument_img2 =
op.MakeArgument(a_device_buf_img.GetDeviceBuffer(),
b_device_buf_re.GetDeviceBuffer(),
std::array<const void*, 1>{e_device_buf_img1.GetDeviceBuffer()},
e_device_buf_img.GetDeviceBuffer(),
......@@ -306,8 +310,6 @@ int run_complex_contraction_bilinear_example(int argc, char* argv[])
b_element_op,
cde_element_op);
if(!op.IsSupportedArgument(argument_img2))
{
std::cout << op.GetTypeString() << " does not support this problem" << std::endl;
......@@ -317,7 +319,6 @@ int run_complex_contraction_bilinear_example(int argc, char* argv[])
float ave_time_img2 = invoker.Run(argument_img2, StreamConfig{nullptr, time_kernel});
ck::index_t M =
ck::accumulate_n<ck::index_t>(e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
......@@ -331,7 +332,7 @@ int run_complex_contraction_bilinear_example(int argc, char* argv[])
std::size_t num_btype = sizeof(ADataType) * M * K + sizeof(BDataType) * K * N +
sizeof(DDataType) * M * N + sizeof(EDataType) * M * N * 2;
float ave_time = ave_time_img2 + ave_time_img1 + ave_time_re2 + ave_time_re1 ;
float ave_time = ave_time_img2 + ave_time_img1 + ave_time_re2 + ave_time_re1;
float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
float gb_per_sec = num_btype / 1.E6 / ave_time;
......@@ -366,8 +367,8 @@ int run_complex_contraction_bilinear_example(int argc, char* argv[])
auto ref_op = ReferenceOpInstance{};
auto ref_invoker = ref_op.MakeInvoker();
auto ref_argument_re =
ref_op.MakeArgument(a_ms_ks_re, b_ns_ks_re, c_ms_ns_host_result_re, a_element_op, b_element_op);
auto ref_argument_re = ref_op.MakeArgument(
a_ms_ks_re, b_ns_ks_re, c_ms_ns_host_result_re, a_element_op, b_element_op);
ref_invoker.Run(ref_argument_re);
......@@ -376,7 +377,6 @@ int run_complex_contraction_bilinear_example(int argc, char* argv[])
cde_element_op = CDEElementOp{alpha, beta};
for(size_t m0 = 0; m0 < e_ms_ns_host_result_re.mDesc.GetLengths()[0]; ++m0)
{
for(size_t m1 = 0; m1 < e_ms_ns_host_result_re.mDesc.GetLengths()[1]; ++m1)
......@@ -398,8 +398,8 @@ int run_complex_contraction_bilinear_example(int argc, char* argv[])
cde_element_op = CDEElementOp{alpha, beta};
auto ref_argument_re1 =
ref_op.MakeArgument(a_ms_ks_img, b_ns_ks_img, c_ms_ns_host_result_re1, a_element_op, b_element_op);
auto ref_argument_re1 = ref_op.MakeArgument(
a_ms_ks_img, b_ns_ks_img, c_ms_ns_host_result_re1, a_element_op, b_element_op);
ref_invoker.Run(ref_argument_re1);
......@@ -421,15 +421,12 @@ int run_complex_contraction_bilinear_example(int argc, char* argv[])
isRealOk = ck::utils::check_err(e_ms_ns_device_result_re, e_ms_ns_host_result_re) ? 0 : 1;
// Img Part Verification
Tensor<CShuffleDataType> c_ms_ns_host_result_img(e_ms_ns_lengths, e_ms_ns_strides);
Tensor<CShuffleDataType> c_ms_ns_host_result_img1(e_ms_ns_lengths, e_ms_ns_strides);
auto ref_argument_img =
ref_op.MakeArgument(a_ms_ks_re, b_ns_ks_img, c_ms_ns_host_result_img, a_element_op, b_element_op);
auto ref_argument_img = ref_op.MakeArgument(
a_ms_ks_re, b_ns_ks_img, c_ms_ns_host_result_img, a_element_op, b_element_op);
ref_invoker.Run(ref_argument_img);
......@@ -454,8 +451,8 @@ int run_complex_contraction_bilinear_example(int argc, char* argv[])
}
}
auto ref_argument_img1 =
ref_op.MakeArgument(a_ms_ks_img, b_ns_ks_re, c_ms_ns_host_result_img1, a_element_op, b_element_op);
auto ref_argument_img1 = ref_op.MakeArgument(
a_ms_ks_img, b_ns_ks_re, c_ms_ns_host_result_img1, a_element_op, b_element_op);
ref_invoker.Run(ref_argument_img1);
......
example/65_complex_contraction_scale/run_complex_contraction_scale_example.inc 100755 → 100644
View file @ 1e5e2dc3
......@@ -107,7 +107,6 @@ int run_complex_contraction_scale_example(int argc, char* argv[])
Tensor<ADataType> a_ms_ks_img(a_ms_ks_lengths, a_ms_ks_strides);
Tensor<BDataType> b_ns_ks_img(b_ns_ks_lengths, b_ns_ks_strides);
Tensor<EDataType> e_ms_ns_host_result_img(e_ms_ns_lengths, e_ms_ns_strides);
Tensor<EDataType> e_ms_ns_device_result_img(e_ms_ns_lengths, e_ms_ns_strides);
......@@ -145,32 +144,38 @@ int run_complex_contraction_scale_example(int argc, char* argv[])
a_ms_ks_img.GenerateTensorValue(GeneratorTensor_3<ADataType>{0.0, 1.0});
b_ns_ks_img.GenerateTensorValue(GeneratorTensor_3<BDataType>{-0.5, 0.5});
break;
}
DeviceMem a_device_buf_re(sizeof(ADataType) * a_ms_ks_re.mDesc.GetElementSpaceSize());
DeviceMem b_device_buf_re(sizeof(BDataType) * b_ns_ks_re.mDesc.GetElementSpaceSize());
DeviceMem e_device_buf_re(sizeof(EDataType) * e_ms_ns_device_result_re.mDesc.GetElementSpaceSize());
DeviceMem e_device_buf_re(sizeof(EDataType) *
e_ms_ns_device_result_re.mDesc.GetElementSpaceSize());
DeviceMem a_device_buf_img(sizeof(ADataType) * a_ms_ks_img.mDesc.GetElementSpaceSize());
DeviceMem b_device_buf_img(sizeof(BDataType) * b_ns_ks_img.mDesc.GetElementSpaceSize());
DeviceMem e_device_buf_img(sizeof(EDataType) * e_ms_ns_device_result_img.mDesc.GetElementSpaceSize());
<<<<<<< HEAD
// Intermediate Value For E Real and Img
DeviceMem e_device_buf_re1(sizeof(EDataType) * e_ms_ns_device_result_re.mDesc.GetElementSpaceSize());
DeviceMem e_device_buf_img1(sizeof(EDataType) * e_ms_ns_device_result_img.mDesc.GetElementSpaceSize());
=======
DeviceMem e_device_buf_img(sizeof(EDataType) *
e_ms_ns_device_result_img.mDesc.GetElementSpaceSize());
// // Intermediate Value For E Real and Img
// DeviceMem e_device_buf_re1(sizeof(EDataType) *
// e_ms_ns_device_result_re.mDesc.GetElementSpaceSize()); DeviceMem
// e_device_buf_img1(sizeof(EDataType) * e_ms_ns_device_result_img.mDesc.GetElementSpaceSize());
>>>>>>> 81eece66cb698622dafbc0f2c726ba743eb345fe
a_device_buf_re.ToDevice(a_ms_ks_re.mData.data());
b_device_buf_re.ToDevice(b_ns_ks_re.mData.data());
a_device_buf_img.ToDevice(a_ms_ks_img.mData.data());
b_device_buf_img.ToDevice(b_ns_ks_img.mData.data());
// set zero
e_device_buf_re.SetZero();
e_device_buf_img.SetZero();
......@@ -178,13 +183,23 @@ int run_complex_contraction_scale_example(int argc, char* argv[])
e_device_buf_img1.SetZero();
<<<<<<< HEAD
auto a_element_op = AElementOp{};
auto b_element_op = BElementOp{};
=======
// // set zero for intermediate values
// e_device_buf_re1.SetZero();
// e_device_buf_img1.SetZero();
auto a_element_op = AElementOp{};
auto b_element_op = BElementOp{};
>>>>>>> 81eece66cb698622dafbc0f2c726ba743eb345fe
auto cde_element_op_scale = CDEElementOp_Scale{scale};
// device operation
// E1_real = A_real * B_real
<<<<<<< HEAD
auto op_scale = DeviceOpInstance_Scale{};
auto invoker_scale = op_scale.MakeInvoker();
auto argument_re1 = op_scale.MakeArgument(a_device_buf_re.GetDeviceBuffer(),
......@@ -202,6 +217,27 @@ int run_complex_contraction_scale_example(int argc, char* argv[])
a_element_op,
b_element_op,
cde_element_op_scale);
=======
auto op = DeviceOpInstance{};
auto invoker = op.MakeInvoker();
auto argument_re1 =
op.MakeArgument(a_device_buf_re.GetDeviceBuffer(),
b_device_buf_re.GetDeviceBuffer(),
// std::array<const void*, 1>{d_device_buf_re.GetDeviceBuffer()},
std::array<const void*, 0>{},
e_device_buf_re.GetDeviceBuffer(),
a_ms_ks_lengths,
a_ms_ks_strides,
b_ns_ks_lengths,
b_ns_ks_strides,
std::array<std::vector<ck::index_t>, 0>{},
std::array<std::vector<ck::index_t>, 0>{},
e_ms_ns_lengths,
e_ms_ns_strides,
a_element_op,
b_element_op,
cde_element_op_scale);
>>>>>>> 81eece66cb698622dafbc0f2c726ba743eb345fe
if(!op_scale.IsSupportedArgument(argument_re1))
{
......@@ -212,10 +248,10 @@ int run_complex_contraction_scale_example(int argc, char* argv[])
float ave_time_re1 = invoker_scale.Run(argument_re1, StreamConfig{nullptr, time_kernel});
alpha = -1.f * scale;
beta = 1.f;
<<<<<<< HEAD
auto cde_element_op = CDEElementOp{alpha, beta};
......@@ -241,6 +277,31 @@ int run_complex_contraction_scale_example(int argc, char* argv[])
a_element_op,
b_element_op,
cde_element_op);
=======
a_element_op = AElementOp{};
b_element_op = BElementOp{};
auto cde_element_op = CDEElementOp{alpha, beta};
// device operation
// For real Intermediate Value re_2
auto argument_re2 =
op.MakeArgument(a_device_buf_img.GetDeviceBuffer(),
b_device_buf_img.GetDeviceBuffer(),
std::array<const void*, 1>{e_device_buf_re.GetDeviceBuffer()},
e_device_buf_re.GetDeviceBuffer(),
a_ms_ks_lengths,
a_ms_ks_strides,
b_ns_ks_lengths,
b_ns_ks_strides,
std::array<std::vector<ck::index_t>, 1>{d_ms_ns_lengths},
std::array<std::vector<ck::index_t>, 1>{d_ms_ns_strides},
e_ms_ns_lengths,
e_ms_ns_strides,
a_element_op,
b_element_op,
cde_element_op);
>>>>>>> 81eece66cb698622dafbc0f2c726ba743eb345fe
if(!op.IsSupportedArgument(argument_re2))
{
......@@ -251,10 +312,23 @@ int run_complex_contraction_scale_example(int argc, char* argv[])
float ave_time_re2 = invoker.Run(argument_re2, StreamConfig{nullptr, time_kernel});
<<<<<<< HEAD
auto argument_img1 = op_scale.MakeArgument(a_device_buf_re.GetDeviceBuffer(),
b_device_buf_img.GetDeviceBuffer(),
std::array<const void*, 0>{},
=======
// scale = 1.f ;
// a_element_op = AElementOp{};
// b_element_op = BElementOp{};
// cde_element_op = CDEElementOp{alpha, beta};
auto argument_img1 = op.MakeArgument(a_device_buf_re.GetDeviceBuffer(),
b_device_buf_img.GetDeviceBuffer(),
std::array<const void*, 0>{}
},
>>>>>>> 81eece66cb698622dafbc0f2c726ba743eb345fe
e_device_buf_img.GetDeviceBuffer(),
a_ms_ks_lengths,
a_ms_ks_strides,
......@@ -268,22 +342,25 @@ int run_complex_contraction_scale_example(int argc, char* argv[])
b_element_op,
cde_element_op_scale);
if(!op.IsSupportedArgument(argument_img1))
{
std::cout << op.GetTypeString() << " does not support this problem" << std::endl;
<<<<<<< HEAD
if(!op_scale.IsSupportedArgument(argument_img1))
{
std::cout << op_scale.GetTypeString() << " does not support this problem" << std::endl;
=======
return 0;
}
float ave_time_img1 = invoker_scale.Run(argument_img1, StreamConfig{nullptr, time_kernel});
alpha = 1.f * scale;
beta = 1.f;
}
>>>>>>> 81eece66cb698622dafbc0f2c726ba743eb345fe
float ave_time_img1 = invoker.Run(argument_img1, StreamConfig{nullptr, time_kernel});
alpha = 1.f * scale;
beta = 1.f;
auto argument_img2 = op.MakeArgument(a_device_buf_img.GetDeviceBuffer(),
auto argument_img2 = op.MakeArgument(a_device_buf_img.GetDeviceBuffer(),
b_device_buf_re.GetDeviceBuffer(),
std::array<const void*, 1>{e_device_buf_img.GetDeviceBuffer()},
e_device_buf_img.GetDeviceBuffer(),
......@@ -299,50 +376,76 @@ int run_complex_contraction_scale_example(int argc, char* argv[])
b_element_op,
cde_element_op);
if(!op.IsSupportedArgument(argument_img2))
{
if(!op.IsSupportedArgument(argument_img2))
{
std::cout << op.GetTypeString() << " does not support this problem" << std::endl;
<<<<<<< HEAD
float ave_time_img1 = invoker_scale.Run(argument_img1, StreamConfig{nullptr, time_kernel});
=======
return 0;
}
}
>>>>>>> 81eece66cb698622dafbc0f2c726ba743eb345fe
float ave_time_img2 = invoker.Run(argument_img2, StreamConfig{nullptr, time_kernel});
float ave_time_img2 = invoker.Run(argument_img2, StreamConfig{nullptr, time_kernel});
<<<<<<< HEAD
ck::index_t M =
auto argument_img2 = op.MakeArgument(a_device_buf_img.GetDeviceBuffer(),
b_device_buf_re.GetDeviceBuffer(),
std::array<const void*, 1>{e_device_buf_img.GetDeviceBuffer()},
e_device_buf_img.GetDeviceBuffer(),
a_ms_ks_lengths,
a_ms_ks_strides,
b_ns_ks_lengths,
b_ns_ks_strides,
std::array<std::vector<ck::index_t>, 1>{d_ms_ns_lengths},
std::array<std::vector<ck::index_t>, 1>{d_ms_ns_strides},
e_ms_ns_lengths,
e_ms_ns_strides,
a_element_op,
b_element_op,
cde_element_op);
=======
ck::index_t M =
ck::accumulate_n<ck::index_t>(e_ms_ns_lengths.begin(), NumDimM, 1, std::multiplies<>{});
>>>>>>> 81eece66cb698622dafbc0f2c726ba743eb345fe
ck::index_t N = ck::accumulate_n<ck::index_t>(
ck::index_t N = ck::accumulate_n<ck::index_t>(
e_ms_ns_lengths.begin() + NumDimM, NumDimN, 1, std::multiplies<>{});
ck::index_t K = ck::accumulate_n<ck::index_t>(
ck::index_t K = ck::accumulate_n<ck::index_t>(
a_ms_ks_lengths.begin() + NumDimM, NumDimK, 1, std::multiplies<>{});
std::size_t flop = std::size_t(2) * M * N * K * 2;
std::size_t num_btype = (sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(EDataType) * M * N) * 2;
std::size_t flop = std::size_t(2) * M * N * K * 2;
std::size_t num_btype = sizeof(ADataType) * M * K + sizeof(BDataType) * K * N +
sizeof(DDataType) * M * N + sizeof(EDataType) * M * N * 2;
float ave_time = ave_time_img2 + ave_time_img1 + ave_time_re2 + ave_time_re1 ;
float ave_time = ave_time_img2 + ave_time_img1 + ave_time_re2 + ave_time_re1;
float tflops = static_cast<float>(flop) / 1.E9 / ave_time;
float gb_per_sec = num_btype / 1.E6 / ave_time;
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::cout << "Perf: " << ave_time << " ms, " << tflops << " TFlops, " << gb_per_sec << " GB/s, "
<< op.GetTypeString() << std::endl;
e_device_buf_re.FromDevice(e_ms_ns_device_result_re.mData.data());
e_device_buf_img.FromDevice(e_ms_ns_device_result_img.mData.data());
e_device_buf_re.FromDevice(e_ms_ns_device_result_re.mData.data());
e_device_buf_img.FromDevice(e_ms_ns_device_result_img.mData.data());
auto isRealOk = 0;
auto isImgOk = 0;
auto isRealOk = 0;
auto isImgOk = 0;
if(do_verification)
{
if(do_verification)
{
// Real Part Verification
Tensor<CShuffleDataType> c_ms_ns_host_result_re(e_ms_ns_lengths, e_ms_ns_strides);
Tensor<CShuffleDataType> c_ms_ns_host_result_re1(e_ms_ns_lengths, e_ms_ns_strides);
<<<<<<< HEAD
std::size_t flop = std::size_t(2) * M * N * K * 2;
std::size_t num_btype = (sizeof(ADataType) * M * K + sizeof(BDataType) * K * N + sizeof(EDataType) * M * N) * 2;
=======
using ReferenceOpInstance =
ck::tensor_operation::host::ReferenceContraction_M2_N2_K2<NumDimM,
NumDimN,
......@@ -354,11 +457,13 @@ int run_complex_contraction_scale_example(int argc, char* argv[])
F32,
AElementOp,
BElementOp>;
>>>>>>> 81eece66cb698622dafbc0f2c726ba743eb345fe
auto ref_op = ReferenceOpInstance{};
auto ref_invoker = ref_op.MakeInvoker();
auto ref_argument_re = ref_op.MakeArgument(a_ms_ks_re, b_ns_ks_re, c_ms_ns_host_result_re, a_element_op, b_element_op);
auto ref_argument_re = ref_op.MakeArgument(
a_ms_ks_re, b_ns_ks_re, c_ms_ns_host_result_re, a_element_op, b_element_op);
ref_invoker.Run(ref_argument_re);
......@@ -367,7 +472,6 @@ int run_complex_contraction_scale_example(int argc, char* argv[])
// cde_element_op = CDEElementOp{alpha, beta};
for(size_t m0 = 0; m0 < e_ms_ns_host_result_re.mDesc.GetLengths()[0]; ++m0)
{
for(size_t m1 = 0; m1 < e_ms_ns_host_result_re.mDesc.GetLengths()[1]; ++m1)
......@@ -388,8 +492,8 @@ int run_complex_contraction_scale_example(int argc, char* argv[])
cde_element_op = CDEElementOp{alpha, beta};
auto ref_argument_re1 =
ref_op.MakeArgument(a_ms_ks_img, b_ns_ks_img, c_ms_ns_host_result_re1, a_element_op, b_element_op);
auto ref_argument_re1 = ref_op.MakeArgument(
a_ms_ks_img, b_ns_ks_img, c_ms_ns_host_result_re1, a_element_op, b_element_op);
ref_invoker.Run(ref_argument_re1);
......@@ -411,15 +515,12 @@ int run_complex_contraction_scale_example(int argc, char* argv[])
isRealOk = ck::utils::check_err(e_ms_ns_device_result_re, e_ms_ns_host_result_re) ? 0 : 1;
// Img Part Verification
Tensor<CShuffleDataType> c_ms_ns_host_result_img(e_ms_ns_lengths, e_ms_ns_strides);
Tensor<CShuffleDataType> c_ms_ns_host_result_img1(e_ms_ns_lengths, e_ms_ns_strides);
auto ref_argument_img =
ref_op.MakeArgument(a_ms_ks_re, b_ns_ks_img, c_ms_ns_host_result_img, a_element_op, b_element_op);
auto ref_argument_img = ref_op.MakeArgument(
a_ms_ks_re, b_ns_ks_img, c_ms_ns_host_result_img, a_element_op, b_element_op);
ref_invoker.Run(ref_argument_img);
......@@ -446,13 +547,11 @@ int run_complex_contraction_scale_example(int argc, char* argv[])
alpha = 1.f * scale;
beta = -1.f;
auto ref_argument_img1 =
ref_op.MakeArgument(a_ms_ks_img, b_ns_ks_re, c_ms_ns_host_result_img1, a_element_op, b_element_op);
auto ref_argument_img1 = ref_op.MakeArgument(
a_ms_ks_img, b_ns_ks_re, c_ms_ns_host_result_img1, a_element_op, b_element_op);
ref_invoker.Run(ref_argument_img1);
for(size_t m0 = 0; m0 < e_ms_ns_host_result_img.mDesc.GetLengths()[0]; ++m0)
{
for(size_t m1 = 0; m1 < e_ms_ns_host_result_img.mDesc.GetLengths()[1]; ++m1)
......@@ -472,7 +571,7 @@ int run_complex_contraction_scale_example(int argc, char* argv[])
isImgOk = ck::utils::check_err(e_ms_ns_device_result_re, e_ms_ns_host_result_re) ? 0 : 1;
return (isRealOk && isImgOk);
}
}
return 0;
return 0;
}
script/clang-format-overwrite.sh
View file @ 1e5e2dc3
#find . -name deps -prune -o -name build -prune -o -iname '*.h' -o -iname '*.hpp' -o -iname '*.cpp' -o -iname '*.h.in' -o -iname '*.hpp.in' -o -iname '*.cpp.in' -o -iname '*.cl' -o -iname '*.cuh' -o -iname '*.cu' -o -iname '*.inc' | xargs -n 1 -P 16 -I{} -t sh -c 'clang-format-12 -i -style=file {}'
find . -name deps -prune -o -name build -prune -o -iname '*.h' -o -iname '*.hpp' -o -iname '*.cpp' -o -iname '*.h.in' -o -iname '*.hpp.in' -o -iname '*.cpp.in' -o -iname '*.cl' -o -iname '*.cuh' -o -iname '*.cu' -o -iname '*.inc' | xargs -n 1 -P 16 -I{} -t sh -c 'clang-format-12 -i -style=file {}'
git status --porcelain | awk '$1 != "D" && (match($2, "\\.cpp|hpp|inc")) {print $2}' | xargs -n 1 -P 16 -I{} -t sh -c 'clang-format-12 -i -style=file {}'
test/complex_contraction_bilinear/test_complex_contraction_bilinear.cpp 100755 → 100644
View file @ 1e5e2dc3
......@@ -12,8 +12,6 @@
#include "profiler/profile_contraction_impl.hpp"
#include "profiler/profile_contraction_utils.hpp"
using F32 = float;
using F64 = double;
......@@ -93,7 +91,6 @@ class TestContraction : public ::testing::Test
}
};
template <typename Tuple>
class TestContractionBilinear : public TestContraction<Tuple>
{
......@@ -109,10 +106,8 @@ using BilinearKernelTypes =
::testing::Types<ALL_LAYOUT_COMBINATIONS(F32, ck::Tuple<F32>, F32, Bilinear),
ALL_LAYOUT_COMBINATIONS(F64, ck::Tuple<F64>, F64, Bilinear)>;
TYPED_TEST_SUITE(TestContractionBilinear, BilinearKernelTypes);
TYPED_TEST(TestContractionBilinear, bilinear)
{
this->p_cd_element_op = std::make_unique<Bilinear>(1.f, 1.f);
......@@ -120,5 +115,3 @@ TYPED_TEST(TestContractionBilinear, bilinear)
this->p_cd_element_op = std::make_unique<Bilinear>(-0.5f, 0.5f);
this->Run();
}
test/complex_contraction_scale/test_complex_contraction_scale.cpp 100755 → 100644
View file @ 1e5e2dc3
......@@ -18,7 +18,6 @@ using F64 = double;
using Row = ck::tensor_layout::gemm::RowMajor;
using Col = ck::tensor_layout::gemm::ColumnMajor;
using Scale = ck::tensor_operation::element_wise::Scale;
struct Dimensions
......@@ -96,24 +95,17 @@ class TestContractionScale : public TestContraction<Tuple>
{
};
#define ALL_LAYOUT_COMBINATIONS(dt, tuple_dt, compute_dt, op) \
std::tuple<Row, Row, Row, dt, tuple_dt, compute_dt, op>, \
std::tuple<Row, Col, Row, dt, tuple_dt, compute_dt, op>, \
std::tuple<Col, Row, Row, dt, tuple_dt, compute_dt, op>, \
std::tuple<Col, Col, Row, dt, tuple_dt, compute_dt, op>
using ScaleKernelTypes = ::testing::Types<ALL_LAYOUT_COMBINATIONS(F32, ck::Tuple<>, F32, Scale),
ALL_LAYOUT_COMBINATIONS(F64, ck::Tuple<>, F64, Scale)>;
TYPED_TEST_SUITE(TestContractionScale, ScaleKernelTypes);
TYPED_TEST(TestContractionScale, scale)
{
this->p_cd_element_op = std::make_unique<Scale>(1.f);
......@@ -121,7 +113,3 @@ TYPED_TEST(TestContractionScale, scale)
this->p_cd_element_op = std::make_unique<Scale>(0.5f);
this->Run();
}
test/complex_contraction_scale/test_complex_contraction_scale_interface.cpp 100755 → 100644
View file @ 1e5e2dc3
......@@ -18,7 +18,6 @@
using Pass = ck::tensor_operation::element_wise::PassThrough;
using Scale = ck::tensor_operation::element_wise::Scale;
template <ck::index_t... Is>
using S = ck::Sequence<Is...>;
......@@ -101,7 +100,8 @@ class ContractionDeviceOpWrapper
{
bool supported = false;
const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<DeviceOp>::GetInstances();
const auto op_ptrs = ck::tensor_operation::device::instance::DeviceOperationInstanceFactory<
DeviceOp>::GetInstances();
for(auto& op_ptr : op_ptrs)
{
......
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