Skip to content

update intel_gpu kernels #2509

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
co63oc wants to merge 2 commits into from
Closed

update intel_gpu kernels #2509

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
6b1a77f
update kernels namespace
co63oc Apr 21, 2026
953d343
fix
co63oc Apr 21, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
30 changes: 15 additions & 15 deletions backends/intel_gpu/kernels/argsort_kernel.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,13 +17,13 @@
#include "kernels/phi_funcs.h"
#include "paddle/phi/capi/all.h"

namespace custom_kernel {
namespace phi {

namespace gpu {

template <typename T>
void Transpose(const phi::Context& ctx,
const phi::DenseTensor& x,
void Transpose(const Context& ctx,
const DenseTensor& x,
const std::vector<int64_t>& axis,
T* out_data,
const std::vector<int64_t>& out_dims,
Expand Down Expand Up @@ -113,12 +113,12 @@ void FullSort(int input_height,
}

template <typename T>
void ArgsortKernel(const phi::Context& dev_ctx,
const phi::DenseTensor& input,
void ArgsortKernel(const Context& dev_ctx,
const DenseTensor& input,
int axis,
bool descending,
phi::DenseTensor* output,
phi::DenseTensor* indices) {
DenseTensor* output,
DenseTensor* indices) {
auto in_dims = input.dims();
auto out_dims = output->dims();
auto out_size = output->numel();
Expand All @@ -142,7 +142,7 @@ void ArgsortKernel(const phi::Context& dev_ctx,
n = in_dims[0];
m = in_dims[1];
}
phi::DenseTensor cpu_input;
DenseTensor cpu_input;
cpu_input.Resize(std::vector<int64_t>(in_dims));
cpu_input.set_dtype(input.dtype());
auto cpu_input_data = dev_ctx.template HostAlloc<T>(&cpu_input);
Expand All @@ -151,14 +151,14 @@ void ArgsortKernel(const phi::Context& dev_ctx,
q->memcpy(cpu_input_data, input_data, input.memory_size());
q->wait();
// cpu implement
phi::DenseTensor cpu_output;
DenseTensor cpu_output;
cpu_output.Resize(std::vector<int64_t>(out_dims));
cpu_output.set_dtype(output->dtype());
auto cpu_output_dims = cpu_output.dims();
auto cpu_output_numel = cpu_output.numel();
auto cpu_output_data = dev_ctx.template HostAlloc<T>(&cpu_output);

phi::DenseTensor cpu_ids;
DenseTensor cpu_ids;
cpu_ids.Resize(std::vector<int64_t>(indices->dims()));
cpu_ids.set_dtype(indices->dtype());
auto cpu_ids_dims = cpu_ids.dims();
Expand Down Expand Up @@ -191,7 +191,7 @@ void ArgsortKernel(const phi::Context& dev_ctx,
trans_dims[i] = in_dims[trans[i]];
}

phi::DenseTensor trans_inp;
DenseTensor trans_inp;
trans_inp.Resize(trans_dims);
auto trans_input_dims = trans_inp.dims();
auto trans_input_numel = trans_inp.numel();
Expand All @@ -207,12 +207,12 @@ void ArgsortKernel(const phi::Context& dev_ctx,
const int64_t input_height = trans_dims[0];
const int64_t input_width = trans_dims[trans_dims.size() - 1];

phi::DenseTensor cpu_tmp_output;
DenseTensor cpu_tmp_output;
cpu_tmp_output.Resize(trans_dims);
cpu_tmp_output.set_dtype(output->dtype());
auto cpu_tmp_output_data = dev_ctx.template HostAlloc<T>(&cpu_tmp_output);

phi::DenseTensor cpu_tmp_ids;
DenseTensor cpu_tmp_ids;
cpu_tmp_ids.Resize(trans_dims);
cpu_tmp_ids.set_dtype(indices->dtype());
auto cpu_tmp_ids_data = dev_ctx.template HostAlloc<int64_t>(&cpu_tmp_ids);
Expand Down Expand Up @@ -243,12 +243,12 @@ void ArgsortKernel(const phi::Context& dev_ctx,

} // namespace gpu

} // namespace custom_kernel
} // namespace phi

PD_BUILD_PHI_KERNEL(argsort,
intel_gpu,
ALL_LAYOUT,
custom_kernel::gpu::ArgsortKernel,
phi::gpu::ArgsortKernel,
float,
double,
int,
Expand Down
24 changes: 12 additions & 12 deletions backends/intel_gpu/kernels/assign_value_kernel.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,14 +16,14 @@
#include "kernels/phi_funcs.h"
#include "paddle/phi/capi/all.h"

namespace custom_kernel {
namespace phi {

template <typename T>
void AssignValueKernel(const phi::Context& dev_ctx,
void AssignValueKernel(const Context& dev_ctx,
const std::vector<int>& shape,
phi::DataType dtype,
const std::vector<phi::Scalar>& values,
phi::DenseTensor* out) {
DenseTensor* out) {
show_kernel("AssignValue-SYCL, type=" << dnn_support::type2String<T>::name());

auto template_dtype = phi::capi::CppTypeToPDType<T>::Type();
Expand All @@ -49,18 +49,18 @@ void AssignValueKernel(const phi::Context& dev_ctx,
}

template <typename T>
void AssignKernel(const phi::Context& dev_ctx,
const phi::DenseTensor& x,
phi::DenseTensor* out) {
void AssignKernel(const Context& dev_ctx,
const DenseTensor& x,
DenseTensor* out) {
auto out_data = dev_ctx.template Alloc<T>(out);
auto x_data = x.data<T>();
std::memcpy(out_data, x_data, sizeof(T) * x.numel());
}

template <typename T>
void AssignRawKernel(const phi::Context& dev_ctx,
const paddle::optional<phi::DenseTensor>& x,
phi::DenseTensor* out) {
void AssignRawKernel(const Context& dev_ctx,
const paddle::optional<DenseTensor>& x,
DenseTensor* out) {
show_kernel("AssignRaw-SYCL, type=" << dnn_support::type2String<T>::name());

if (x) {
Expand All @@ -76,12 +76,12 @@ void AssignRawKernel(const phi::Context& dev_ctx,
}
}

} // namespace custom_kernel
} // namespace phi

PD_BUILD_PHI_KERNEL(assign_value,
intel_gpu,
ALL_LAYOUT,
custom_kernel::AssignValueKernel,
phi::AssignValueKernel,
int,
int64_t,
float,
Expand All @@ -90,7 +90,7 @@ PD_BUILD_PHI_KERNEL(assign_value,
PD_BUILD_PHI_KERNEL(assign_raw,
intel_gpu,
ALL_LAYOUT,
custom_kernel::AssignRawKernel,
phi::AssignRawKernel,
int,
int64_t,
float,
Expand Down
12 changes: 6 additions & 6 deletions backends/intel_gpu/kernels/cast_kernel.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,13 +16,13 @@
#include "kernels/phi_funcs.h"
#include "paddle/phi/capi/all.h"

namespace custom_kernel {
namespace phi {

template <typename T>
void CastKernel(const phi::Context& dev_ctx,
const phi::DenseTensor& x,
void CastKernel(const Context& dev_ctx,
const DenseTensor& x,
phi::DataType out_dtype,
phi::DenseTensor* out) {
DenseTensor* out) {
show_kernel("Cast-SYCL");

auto x_data = x.data<T>();
Expand Down Expand Up @@ -109,12 +109,12 @@ void CastKernel(const phi::Context& dev_ctx,
q->wait();
}

} // namespace custom_kernel
} // namespace phi

PD_BUILD_PHI_KERNEL(cast,
intel_gpu,
ALL_LAYOUT,
custom_kernel::CastKernel,
phi::CastKernel,
float,
double,
int,
Expand Down
86 changes: 43 additions & 43 deletions backends/intel_gpu/kernels/compare_kernel.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,15 +16,15 @@
#include "kernels/phi_funcs.h"
#include "paddle/phi/capi/all.h"

namespace custom_kernel {
namespace phi {

template <typename T, typename F, typename FF>
void RawCompareKernelSycl(const phi::Context& dev_ctx,
void RawCompareKernelSycl(const Context& dev_ctx,
std::string kernel_name,
const phi::DenseTensor& x,
const phi::DenseTensor& y,
const DenseTensor& x,
const DenseTensor& y,
int axis,
phi::DenseTensor* out,
DenseTensor* out,
const F& func,
const FF& float_func) {
show_kernel(kernel_name << "-SYCL type="
Expand All @@ -51,13 +51,13 @@ void RawCompareKernelSycl(const phi::Context& dev_ctx,
}

template <typename T>
void RawCompareKernelDNN(const phi::Context& dev_ctx,
void RawCompareKernelDNN(const Context& dev_ctx,
std::string kernel_name,
dnnl::algorithm binary_type,
const phi::DenseTensor& x,
const phi::DenseTensor& y,
const DenseTensor& x,
const DenseTensor& y,
int axis,
phi::DenseTensor* out) {
DenseTensor* out) {
show_kernel(kernel_name << "-DNN type="
<< dnn_support::type2String<T>::name());

Expand Down Expand Up @@ -106,13 +106,13 @@ void RawCompareKernelDNN(const phi::Context& dev_ctx,
}

template <typename T, typename F, typename FF>
void EqualityKernel(const phi::Context& dev_ctx,
void EqualityKernel(const Context& dev_ctx,
std::string kernel_name,
dnnl::algorithm binary_type,
const phi::DenseTensor& x,
const phi::DenseTensor& y,
const DenseTensor& x,
const DenseTensor& y,
int axis,
phi::DenseTensor* out,
DenseTensor* out,
const F& func,
const FF& float_func) {
if constexpr (std::is_same<T, float>::value) {
Expand All @@ -124,13 +124,13 @@ void EqualityKernel(const phi::Context& dev_ctx,
}

template <typename T, typename F>
void CompareKernel(const phi::Context& dev_ctx,
void CompareKernel(const Context& dev_ctx,
std::string kernel_name,
dnnl::algorithm binary_type,
const phi::DenseTensor& x,
const phi::DenseTensor& y,
const DenseTensor& x,
const DenseTensor& y,
int axis,
phi::DenseTensor* out,
DenseTensor* out,
const F& func) {
if constexpr (std::is_same<T, float>::value) {
RawCompareKernelDNN<T>(dev_ctx, kernel_name, binary_type, x, y, axis, out);
Expand All @@ -140,11 +140,11 @@ void CompareKernel(const phi::Context& dev_ctx,
}

template <typename T>
void NotEqualKernel(const phi::Context& dev_ctx,
const phi::DenseTensor& x,
const phi::DenseTensor& y,
void NotEqualKernel(const Context& dev_ctx,
const DenseTensor& x,
const DenseTensor& y,
int axis,
phi::DenseTensor* out) {
DenseTensor* out) {
EqualityKernel<T>(
dev_ctx,
"NotEqual",
Expand All @@ -163,11 +163,11 @@ void NotEqualKernel(const phi::Context& dev_ctx,
}

template <typename T>
void EqualKernel(const phi::Context& dev_ctx,
const phi::DenseTensor& x,
const phi::DenseTensor& y,
void EqualKernel(const Context& dev_ctx,
const DenseTensor& x,
const DenseTensor& y,
int axis,
phi::DenseTensor* out) {
DenseTensor* out) {
EqualityKernel<T>(
dev_ctx,
"Equal",
Expand All @@ -186,11 +186,11 @@ void EqualKernel(const phi::Context& dev_ctx,
}

template <typename T>
void LessThanKernel(const phi::Context& dev_ctx,
const phi::DenseTensor& x,
const phi::DenseTensor& y,
void LessThanKernel(const Context& dev_ctx,
const DenseTensor& x,
const DenseTensor& y,
int axis,
phi::DenseTensor* out) {
DenseTensor* out) {
CompareKernel<T>(dev_ctx,
"LessThanKernel",
dnnl::algorithm::binary_lt,
Expand All @@ -204,11 +204,11 @@ void LessThanKernel(const phi::Context& dev_ctx,
}

template <typename T>
void LessEqualKernel(const phi::Context& dev_ctx,
const phi::DenseTensor& x,
const phi::DenseTensor& y,
void LessEqualKernel(const Context& dev_ctx,
const DenseTensor& x,
const DenseTensor& y,
int axis,
phi::DenseTensor* out) {
DenseTensor* out) {
CompareKernel<T>(dev_ctx,
"LessEqual",
dnnl::algorithm::binary_le,
Expand All @@ -222,11 +222,11 @@ void LessEqualKernel(const phi::Context& dev_ctx,
}

template <typename T>
void GreaterThanKernel(const phi::Context& dev_ctx,
const phi::DenseTensor& x,
const phi::DenseTensor& y,
void GreaterThanKernel(const Context& dev_ctx,
const DenseTensor& x,
const DenseTensor& y,
int axis,
phi::DenseTensor* out) {
DenseTensor* out) {
CompareKernel<T>(dev_ctx,
"GreaterThan",
dnnl::algorithm::binary_gt,
Expand All @@ -240,11 +240,11 @@ void GreaterThanKernel(const phi::Context& dev_ctx,
}

template <typename T>
void GreaterEqualKernel(const phi::Context& dev_ctx,
const phi::DenseTensor& x,
const phi::DenseTensor& y,
void GreaterEqualKernel(const Context& dev_ctx,
const DenseTensor& x,
const DenseTensor& y,
int axis,
phi::DenseTensor* out) {
DenseTensor* out) {
CompareKernel<T>(dev_ctx,
"GreaterEqual",
dnnl::algorithm::binary_ge,
Expand All @@ -257,13 +257,13 @@ void GreaterEqualKernel(const phi::Context& dev_ctx,
});
}

} // namespace custom_kernel
} // namespace phi

#define PD_REGISTER_COMPARE_KERNEL(name, func) \
PD_BUILD_PHI_KERNEL(name, \
intel_gpu, \
ALL_LAYOUT, \
custom_kernel::func##Kernel, \
phi::func##Kernel, \
float, \
double, \
uint8_t, \
Expand Down
Loading
Loading