Skip to content

Commit

Permalink
Merge branch 'elu_vk_op' of github.com:Yoh-Z/ncnn into elu_vk_op
Browse files Browse the repository at this point in the history
  • Loading branch information
Yoh-Z committed Oct 18, 2022
2 parents 2f29e06 + 7377e98 commit 12593f1
Show file tree
Hide file tree
Showing 2 changed files with 154 additions and 154 deletions.
274 changes: 137 additions & 137 deletions src/layer/vulkan/elu_vulkan.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -18,165 +18,165 @@

namespace ncnn {

ELU_vulkan::ELU_vulkan()
ELU_vulkan::ELU_vulkan()
{
support_vulkan = true;
support_image_storage = true;

pipeline_elu = 0;
pipeline_elu_pack4 = 0;
pipeline_elu_pack8 = 0;
}

int ELU_vulkan::create_pipeline(const Option& opt)
{
const Mat& shape = top_shapes.empty() ? Mat() : top_shapes[0];

int elempack = 1;
if (shape.dims == 1) elempack = opt.use_shader_pack8 && shape.w % 8 == 0 ? 8 : shape.w % 4 == 0 ? 4 : 1;
if (shape.dims == 2) elempack = opt.use_shader_pack8 && shape.h % 8 == 0 ? 8 : shape.h % 4 == 0 ? 4 : 1;
if (shape.dims == 3 || shape.dims == 4) elempack = opt.use_shader_pack8 && shape.c % 8 == 0 ? 8 : shape.c % 4 == 0 ? 4 : 1;

size_t elemsize;
if (opt.use_fp16_storage)
{
support_vulkan = true;
support_image_storage = true;
elemsize = elempack * 2u;
}
else if (opt.use_fp16_packed)
{
elemsize = elempack == 1 ? 4u : elempack * 2u;
}
else
{
elemsize = elempack * 4u;
}

pipeline_elu = 0;
pipeline_elu_pack4 = 0;
pipeline_elu_pack8 = 0;
Mat shape_packed;
if (shape.dims == 1) shape_packed = Mat(shape.w / elempack, (void*)0, elemsize, elempack);
if (shape.dims == 2) shape_packed = Mat(shape.w, shape.h / elempack, (void*)0, elemsize, elempack);
if (shape.dims == 3) shape_packed = Mat(shape.w, shape.h, shape.c / elempack, (void*)0, elemsize, elempack);
if (shape.dims == 4) shape_packed = Mat(shape.w, shape.h, shape.d, shape.c / elempack, (void*)0, elemsize, elempack);

std::vector<vk_specialization_type> specializations(1 + 5);
specializations[0].f = alpha;
specializations[1 + 0].i = shape_packed.dims;
specializations[1 + 1].i = shape_packed.w;
specializations[1 + 2].i = shape_packed.h * shape_packed.d;
specializations[1 + 3].i = shape_packed.c;
specializations[1 + 4].i = shape_packed.cstep;

Mat local_size_xyz;
if (shape_packed.dims == 1)
{
local_size_xyz.w = std::min(64, shape_packed.w);
local_size_xyz.h = 1;
local_size_xyz.c = 1;
}
if (shape_packed.dims == 2)
{
local_size_xyz.w = std::min(8, shape_packed.w);
local_size_xyz.h = std::min(8, shape_packed.h);
local_size_xyz.c = 1;
}
if (shape_packed.dims == 3)
{
local_size_xyz.w = std::min(4, shape_packed.w);
local_size_xyz.h = std::min(4, shape_packed.h);
local_size_xyz.c = std::min(4, shape_packed.c);
}
if (shape_packed.dims == 4)
{
local_size_xyz.w = std::min(4, shape_packed.w);
local_size_xyz.h = std::min(4, shape_packed.h * shape_packed.d);
local_size_xyz.c = std::min(4, shape_packed.c);
}

int ELU_vulkan::create_pipeline(const Option& opt)
// pack1
if (shape.dims == 0 || elempack == 1)
{
const Mat& shape = top_shapes.empty() ? Mat() : top_shapes[0];

int elempack = 1;
if (shape.dims == 1) elempack = opt.use_shader_pack8 && shape.w % 8 == 0 ? 8 : shape.w % 4 == 0 ? 4 : 1;
if (shape.dims == 2) elempack = opt.use_shader_pack8 && shape.h % 8 == 0 ? 8 : shape.h % 4 == 0 ? 4 : 1;
if (shape.dims == 3 || shape.dims == 4) elempack = opt.use_shader_pack8 && shape.c % 8 == 0 ? 8 : shape.c % 4 == 0 ? 4 : 1;

size_t elemsize;
if (opt.use_fp16_storage)
{
elemsize = elempack * 2u;
}
else if (opt.use_fp16_packed)
{
elemsize = elempack == 1 ? 4u : elempack * 2u;
}
else
{
elemsize = elempack * 4u;
}

Mat shape_packed;
if (shape.dims == 1) shape_packed = Mat(shape.w / elempack, (void*)0, elemsize, elempack);
if (shape.dims == 2) shape_packed = Mat(shape.w, shape.h / elempack, (void*)0, elemsize, elempack);
if (shape.dims == 3) shape_packed = Mat(shape.w, shape.h, shape.c / elempack, (void*)0, elemsize, elempack);
if (shape.dims == 4) shape_packed = Mat(shape.w, shape.h, shape.d, shape.c / elempack, (void*)0, elemsize, elempack);

std::vector<vk_specialization_type> specializations(1 + 5);
specializations[0].f = alpha;
specializations[1 + 0].i = shape_packed.dims;
specializations[1 + 1].i = shape_packed.w;
specializations[1 + 2].i = shape_packed.h * shape_packed.d;
specializations[1 + 3].i = shape_packed.c;
specializations[1 + 4].i = shape_packed.cstep;

Mat local_size_xyz;
if (shape_packed.dims == 1)
{
local_size_xyz.w = std::min(64, shape_packed.w);
local_size_xyz.h = 1;
local_size_xyz.c = 1;
}
if (shape_packed.dims == 2)
{
local_size_xyz.w = std::min(8, shape_packed.w);
local_size_xyz.h = std::min(8, shape_packed.h);
local_size_xyz.c = 1;
}
if (shape_packed.dims == 3)
{
local_size_xyz.w = std::min(4, shape_packed.w);
local_size_xyz.h = std::min(4, shape_packed.h);
local_size_xyz.c = std::min(4, shape_packed.c);
}
if (shape_packed.dims == 4)
{
local_size_xyz.w = std::min(4, shape_packed.w);
local_size_xyz.h = std::min(4, shape_packed.h * shape_packed.d);
local_size_xyz.c = std::min(4, shape_packed.c);
}

// pack1
if (shape.dims == 0 || elempack == 1)
{
pipeline_elu = new Pipeline(vkdev);
pipeline_elu->set_optimal_local_size_xyz(local_size_xyz);
pipeline_elu->create(LayerShaderType::elu, opt, specializations);
}

// pack4
if (shape.dims == 0 || elempack == 4)
{
pipeline_elu_pack4 = new Pipeline(vkdev);
pipeline_elu_pack4->set_optimal_local_size_xyz(local_size_xyz);
pipeline_elu_pack4->create(LayerShaderType::elu_pack4, opt, specializations);
}

// pack8
if ((opt.use_shader_pack8 && shape.dims == 0) || elempack == 8)
{
pipeline_elu_pack8 = new Pipeline(vkdev);
pipeline_elu_pack8->set_optimal_local_size_xyz(local_size_xyz);
pipeline_elu_pack8->create(LayerShaderType::elu_pack8, opt, specializations);
}

return 0;
pipeline_elu = new Pipeline(vkdev);
pipeline_elu->set_optimal_local_size_xyz(local_size_xyz);
pipeline_elu->create(LayerShaderType::elu, opt, specializations);
}

int ELU_vulkan::destroy_pipeline(const Option& /*opt*/)
// pack4
if (shape.dims == 0 || elempack == 4)
{
delete pipeline_elu;
pipeline_elu = 0;
pipeline_elu_pack4 = new Pipeline(vkdev);
pipeline_elu_pack4->set_optimal_local_size_xyz(local_size_xyz);
pipeline_elu_pack4->create(LayerShaderType::elu_pack4, opt, specializations);
}

delete pipeline_elu_pack4;
pipeline_elu_pack4 = 0;
// pack8
if ((opt.use_shader_pack8 && shape.dims == 0) || elempack == 8)
{
pipeline_elu_pack8 = new Pipeline(vkdev);
pipeline_elu_pack8->set_optimal_local_size_xyz(local_size_xyz);
pipeline_elu_pack8->create(LayerShaderType::elu_pack8, opt, specializations);
}

delete pipeline_elu_pack8;
pipeline_elu_pack8 = 0;
return 0;
}

return 0;
}
int ELU_vulkan::destroy_pipeline(const Option& /*opt*/)
{
delete pipeline_elu;
pipeline_elu = 0;

int ELU_vulkan::forward_inplace(VkMat& bottom_top_blob, VkCompute& cmd, const Option& /*opt*/) const
{
int elempack = bottom_top_blob.elempack;
delete pipeline_elu_pack4;
pipeline_elu_pack4 = 0;

std::vector<VkMat> bindings(1);
bindings[0] = bottom_top_blob;
delete pipeline_elu_pack8;
pipeline_elu_pack8 = 0;

std::vector<vk_constant_type> constants(5);
constants[0].i = bottom_top_blob.dims;
constants[1].i = bottom_top_blob.w;
constants[2].i = bottom_top_blob.h * bottom_top_blob.d;
constants[3].i = bottom_top_blob.c;
constants[4].i = bottom_top_blob.cstep;
return 0;
}

const Pipeline* pipeline = elempack == 8 ? pipeline_elu_pack8
: elempack == 4 ? pipeline_elu_pack4
: pipeline_elu;
int ELU_vulkan::forward_inplace(VkMat& bottom_top_blob, VkCompute& cmd, const Option& /*opt*/) const
{
int elempack = bottom_top_blob.elempack;

cmd.record_pipeline(pipeline, bindings, constants, bottom_top_blob);
std::vector<VkMat> bindings(1);
bindings[0] = bottom_top_blob;

return 0;
}
std::vector<vk_constant_type> constants(5);
constants[0].i = bottom_top_blob.dims;
constants[1].i = bottom_top_blob.w;
constants[2].i = bottom_top_blob.h * bottom_top_blob.d;
constants[3].i = bottom_top_blob.c;
constants[4].i = bottom_top_blob.cstep;

int ELU_vulkan::forward_inplace(VkImageMat& bottom_top_blob, VkCompute& cmd, const Option& /*opt*/) const
{
int elempack = bottom_top_blob.elempack;
const Pipeline* pipeline = elempack == 8 ? pipeline_elu_pack8
: elempack == 4 ? pipeline_elu_pack4
: pipeline_elu;

std::vector<VkImageMat> bindings(2);
bindings[0] = bottom_top_blob;
bindings[1] = bottom_top_blob;
cmd.record_pipeline(pipeline, bindings, constants, bottom_top_blob);

std::vector<vk_constant_type> constants(5);
constants[0].i = bottom_top_blob.dims;
constants[1].i = bottom_top_blob.w;
constants[2].i = bottom_top_blob.h * bottom_top_blob.d;
constants[3].i = bottom_top_blob.c;
constants[4].i = 0; //bottom_top_blob.cstep;
return 0;
}

const Pipeline* pipeline = elempack == 8 ? pipeline_elu_pack8
: elempack == 4 ? pipeline_elu_pack4
: pipeline_elu;
int ELU_vulkan::forward_inplace(VkImageMat& bottom_top_blob, VkCompute& cmd, const Option& /*opt*/) const
{
int elempack = bottom_top_blob.elempack;

cmd.record_pipeline(pipeline, bindings, constants, bottom_top_blob);
std::vector<VkImageMat> bindings(2);
bindings[0] = bottom_top_blob;
bindings[1] = bottom_top_blob;

return 0;
}
std::vector<vk_constant_type> constants(5);
constants[0].i = bottom_top_blob.dims;
constants[1].i = bottom_top_blob.w;
constants[2].i = bottom_top_blob.h * bottom_top_blob.d;
constants[3].i = bottom_top_blob.c;
constants[4].i = 0; //bottom_top_blob.cstep;

const Pipeline* pipeline = elempack == 8 ? pipeline_elu_pack8
: elempack == 4 ? pipeline_elu_pack4
: pipeline_elu;

cmd.record_pipeline(pipeline, bindings, constants, bottom_top_blob);

return 0;
}

} // namespace ncnn
34 changes: 17 additions & 17 deletions src/layer/vulkan/elu_vulkan.h
Original file line number Diff line number Diff line change
Expand Up @@ -19,23 +19,23 @@

namespace ncnn {

class ELU_vulkan : virtual public ELU
{
public:
ELU_vulkan();

virtual int create_pipeline(const Option& opt);
virtual int destroy_pipeline(const Option& opt);

using ELU::forward_inplace;
virtual int forward_inplace(VkMat& bottom_top_blob, VkCompute& cmd, const Option& opt) const;
virtual int forward_inplace(VkImageMat& bottom_top_blob, VkCompute& cmd, const Option& opt) const;

public:
Pipeline* pipeline_elu;
Pipeline* pipeline_elu_pack4;
Pipeline* pipeline_elu_pack8;
};
class ELU_vulkan : virtual public ELU
{
public:
ELU_vulkan();

virtual int create_pipeline(const Option& opt);
virtual int destroy_pipeline(const Option& opt);

using ELU::forward_inplace;
virtual int forward_inplace(VkMat& bottom_top_blob, VkCompute& cmd, const Option& opt) const;
virtual int forward_inplace(VkImageMat& bottom_top_blob, VkCompute& cmd, const Option& opt) const;

public:
Pipeline* pipeline_elu;
Pipeline* pipeline_elu_pack4;
Pipeline* pipeline_elu_pack8;
};

} // namespace ncnn

Expand Down

0 comments on commit 12593f1

Please sign in to comment.