Skip to content
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

Jump to bottom

[OpenCL]remove process_type to fix FP16 result error #10451

Open
wants to merge 1 commit into
base: develop
Choose a base branch
from
Open

[OpenCL]remove process_type to fix FP16 result error #10451

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
12 changes: 4 additions & 8 deletions lite/kernels/opencl/io_copy_buffer_compute.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -89,8 +89,7 @@ class IoCopyHostToOpenCLCompute
}
#endif
void PrepareForRun() override {
auto& param = Param<param_t>();
if (fp16_support_ && param.process_type != 2) {
if (fp16_support_) {
VLOG(1) << "kernel_func_name_:" << kernel_func_name_;
auto& context = ctx_->As<OpenCLContext>();
context.cl_context()->AddKernel(kernel_func_name_,
Expand All @@ -114,8 +113,7 @@ class IoCopyHostToOpenCLCompute
VLOG(2) << "param.y->dims().size():" << param.y->dims().size();
VLOG(2) << "param.y->dims():" << param.y->dims();
#endif
if (fp16_support_ && param.x->precision() == PRECISION(kFloat) &&
param.process_type != 2) {
if (fp16_support_ && param.x->precision() == PRECISION(kFloat)) {
std::unique_ptr<Tensor> precision_cast_t =
std::unique_ptr<Tensor>(new Tensor);
precision_cast_t->Resize(param.x->dims());
Expand All @@ -132,7 +130,6 @@ class IoCopyHostToOpenCLCompute
kernel_key << kernel_func_name_ << build_options_ << time_stamp_;
auto kernel = context.cl_context()->GetKernel(kernel_key.str());
size_t count = param.x->dims().production();

auto* y_data = MUTABLE_BUFFER_GPU(param.y);
int arg_idx = 0;
cl_int status;
Expand Down Expand Up @@ -204,8 +201,7 @@ class IoCopykOpenCLToHostCompute
}
#endif
void PrepareForRun() override {
auto& param = Param<param_t>();
if (fp16_support_ && param.process_type != 2) {
if (fp16_support_) {
VLOG(1) << "kernel_func_name_:" << kernel_func_name_;
auto& context = ctx_->As<OpenCLContext>();
context.cl_context()->AddKernel(kernel_func_name_,
Expand Down Expand Up @@ -244,7 +240,7 @@ class IoCopykOpenCLToHostCompute
VLOG(4) << "--- Find the sync event for the target cl tensor. ---";
#endif
if (fp16_support_ && param.x->precision() != PRECISION(kInt64) &&
param.x->precision() != PRECISION(kInt32) && param.process_type != 2) {
param.x->precision() != PRECISION(kInt32)) {
mem_size = param.x->dims().production() * sizeof(float);
std::unique_ptr<Tensor> precision_cast_t =
std::unique_ptr<Tensor>(new Tensor);
Expand Down
20 changes: 12 additions & 8 deletions lite/kernels/opencl/layout_image_compute.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,7 +46,7 @@ class LayoutComputeBufferChwToImageDefault
kernel_func_name_ = "buffer_to_image2d_with_pre255";
}
VLOG(1) << "kernel_func_name_:" << kernel_func_name_;
if (param.process_type != 2 && fp16_support_) {
if (fp16_support_) {
build_options_ += " -DMUTABLE_TYPE=half ";
} else {
build_options_ += " -DMUTABLE_TYPE=float ";
Expand Down Expand Up @@ -74,6 +74,10 @@ class LayoutComputeBufferChwToImageDefault
} else {
x_data = param.x->data<float, cl::Buffer>();
}
// std::vector<float> temp(2);
// std::cout << x_data->get() << std::endl;
// TargetWrapperCL::MemcpySync(temp.data(), x_data, 8, IoDirection::DtoH);
// std::cout << "debug layout " << temp[0] << "/" << temp[1] << std::endl;
auto x_dims = param.x->dims();
auto image_shape = InitImageDimInfoWith(x_dims);
auto* y_data = MUTABLE_DATA_GPU(
Expand Down Expand Up @@ -189,7 +193,7 @@ class LayoutComputeImageDefaultToBufferChw
if (param.process_type == 1) {
kernel_func_name_ = "image2d_to_buffer_with_post255";
}
if (param.process_type != 2 && fp16_support_) {
if (fp16_support_) {
build_options_ += " -DMUTABLE_TYPE=half ";
} else {
build_options_ += " -DMUTABLE_TYPE=float ";
Expand Down Expand Up @@ -217,10 +221,10 @@ class LayoutComputeImageDefaultToBufferChw
y_data = param.y->mutable_data<uint8_t, cl::Buffer>(TARGET(kOpenCL));
param.y->set_precision(PRECISION(kInt8));
} else {
y_data = (fp16_support_ && param.process_type != 2)
y_data = (fp16_support_)
? param.y->mutable_data<half_t, cl::Buffer>(TARGET(kOpenCL))
: param.y->mutable_data<float, cl::Buffer>(TARGET(kOpenCL));
if (fp16_support_ && param.process_type != 2)
if (fp16_support_)
param.y->set_precision(PRECISION(kFP16));
else
param.y->set_precision(PRECISION(kFloat));
Expand Down Expand Up @@ -637,7 +641,7 @@ class LayoutComputeImageFolderToBufferChw
if (x_dims.size() > 2) {
kernel_func_name_ = "image2d_to_buffer";
}
if (param.process_type != 2 && fp16_support_) {
if (fp16_support_) {
build_options_ += " -DMUTABLE_TYPE=half ";
} else {
build_options_ += " -DMUTABLE_TYPE=float ";
Expand Down Expand Up @@ -672,11 +676,11 @@ class LayoutComputeImageFolderToBufferChw
x_image_shape = folder_converter.InitImageDimInfoWith(x_dims);
}
auto* y_data =
(fp16_support_ && param.process_type != 2)
(fp16_support_)
? param.y->mutable_data<half_t, cl::Buffer>(TARGET(kOpenCL))
: param.y->mutable_data<float, cl::Buffer>(TARGET(kOpenCL));
auto* x_data = GET_DATA_GPU(param.x);
if (fp16_support_ && param.process_type != 2)
if (fp16_support_)
param.y->set_precision(PRECISION(kFP16));
else
param.y->set_precision(PRECISION(kFloat));
Expand Down Expand Up @@ -787,7 +791,7 @@ class LayoutComputeBufferChwToImageFolder
if (x_dims.size() > 2) {
kernel_func_name_ = "buffer_to_image2d";
}
if (param.process_type != 2 && fp16_support_) {
if (fp16_support_) {
build_options_ += " -DMUTABLE_TYPE=half ";
} else {
build_options_ += " -DMUTABLE_TYPE=float ";
Expand Down