 [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago  [WIP] vulkan compute (#618)
* vulkan infrastructure
* vkallocator and vkmat
* layer interface for vulkan compute
* wip...
* default vulkan device, command wrapper, upload model weight in load_model to simplify layer interface
* simplify command api, vkmat holds staging buffer, relu works
* initialize specialization constant, simplify command dispatch, fix staging buffer copy with different shape, convolution works
* init extension functions
* dynamic local size and group count
* group count=1 is invalid
* regard device max workgroup size limit
* fix relu oooops
* decouple command record and staging allocation
* create result blob
* add pooling shader
* buffer is faster than image :)
* fix pooling shader
* add innerproduct shader
* readonly writeonly decoration
* simplify buffer creation
* decouple command and layer, VK_KHR_descriptor_update_template extension makes descriptor binding update easy :D
* fix vulkan building issues in visual studio (#1)
* fix building issues on visual studio
* ignore benchmark
* cancel changes
* ... ...
* decouple paramdict and vulkandevice
* fix staging buffer destroy in model loading
* remove vkdev member in option
* add padding shader
* simplify vulkan layer creation, simplify convolution and pooling shader for no padding, less debug output
* add convolutiondepthwise and softmax shader
* specialization float type, add leakyrelu
* add dropout shader
* add batchnorm shader
* split vulkan forward
* add scale shader
* push constant type can be int or float
* set_optimal_local_size_xyz
* add eltwise shader
* concat vulkan forward
* fix convolution without bias
* add dummy shader for concat and split, more fix ...
* optional VK_KHR_descriptor_update_template and VK_KHR_push_descriptor
* check VK_KHR_push_descriptor for vkCmdPushDescriptorSetWithTemplateKHR
* binaryop and unaryop shader
* hide raw command buffer
* simple vkbenchncnn benchmark
* create device with transfer queue
* rename command to vkcompute, add vktransfer and layer upload_model interface
* external VkMat, copy and map wrt buffer offset
* command copy respect offset and size
* decouple weight upload and load, simplify upload weight api, use one big staging buffer for uploading weights
* fix build on android
* binding count can not vary :(
* barrier check state, fix sub-op destruction
* declare local_size_xyz constant, fix crash on radv
* fix local_size_xyz, second try
* more barrier and state fix
* fix softmax
* reconstruct buffer memory allocator, reuse blob buffer, less verbose output
* find unified memory type index
* weight staging buffer allocator and weight buffer allocator, respect descriptor buffer offset alignment
* use VK_KHR_descriptor_update_template for faster descriptor update if available, multithread pipeline creation
* find more useful vulkan extensions and enable them
* fix msvc build
* respect VK_KHR_dedicated_allocation for weight buffer allocation
* fix android build
* fix bias name conflicts with metal
* decouple pipeline and layer, building shader sources into shader module, dedicated create_pipeline api, simplify pipeline recording
* drop dummy shader, inplace softmax, multiple shader module works
* fix unique queue family index error
* flatten support vulkan
* mnasnet run
* find shader module by name, each entry point per shader module, fix attribute/id conflict on moltenvk
* some minor changes
* add some high level api
* use dedicated transfer queue to upload weight model
* prefer mappable buffer on unified memory
* global pooling and convolution fc, reuse staging buffer
* implement ring-buffer style blob allocator, add VkBufferMemory capacity
* use blob allocator for workspace blob, it works fine :)
* vulkan option off
* Update layer.cpp
* fix build with vulkan off
* less verbose output, fix crash on vulkan_compute off
* merge benchncnn tool
* allocator clear api, use new weight buffer allocator per net
* add default locked allocator
* mapped mat ptr api, persistent mapped memory works generally :)
* travis ci linux vulkan
* travis ci vulkan wip ...
* more gpu wip ...
* more gpu wip ...
* wip...
* wip...
* wip... ...
* wip... ios vulkan build...
* find glslangValidator on ios build
* use dynamic moltenvk library
* travis ci wip ...
* ios simulator does not support metal at all
* fix cpu only extractor
* optimize workgroup size, first try
* optimize workgroup size, second try
* conv1x1s1d1 vec4
* revert build system
* fix ncnn2mem build
* fix ncnn2mem build
7 years ago |
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731 |
- // Tencent is pleased to support the open source community by making ncnn available.
- //
- // Copyright (C) 2017 THL A29 Limited, a Tencent company. All rights reserved.
- //
- // Licensed under the BSD 3-Clause License (the "License"); you may not use this file except
- // in compliance with the License. You may obtain a copy of the License at
- //
- // https://opensource.org/licenses/BSD-3-Clause
- //
- // Unless required by applicable law or agreed to in writing, software distributed
- // under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR
- // CONDITIONS OF ANY KIND, either express or implied. See the License for the
- // specific language governing permissions and limitations under the License.
-
- #include "concat.h"
- #include <algorithm>
-
- namespace ncnn {
-
- DEFINE_LAYER_CREATOR(Concat)
-
- Concat::Concat()
- {
- one_blob_only = false;
- support_inplace = false;
- support_vulkan = true;
-
- #if NCNN_VULKAN
- pipeline_concat = 0;
- pipeline_concat_pack4 = 0;
- pipeline_concat_pack4to1 = 0;
- #endif // NCNN_VULKAN
- }
-
- int Concat::load_param(const ParamDict& pd)
- {
- axis = pd.get(0, 0);
-
- return 0;
- }
-
- int Concat::forward(const std::vector<Mat>& bottom_blobs, std::vector<Mat>& top_blobs, const Option& opt) const
- {
- int dims = bottom_blobs[0].dims;
- size_t elemsize = bottom_blobs[0].elemsize;
-
- if (dims == 1) // axis == 0
- {
- // concat vector
- // total length
- int top_w = 0;
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const Mat& bottom_blob = bottom_blobs[b];
- top_w += bottom_blob.w;
- }
-
- Mat& top_blob = top_blobs[0];
- top_blob.create(top_w, elemsize, opt.blob_allocator);
- if (top_blob.empty())
- return -100;
-
- float* outptr = top_blob;
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const Mat& bottom_blob = bottom_blobs[b];
-
- int w = bottom_blob.w;
-
- const float* ptr = bottom_blob;
- memcpy(outptr, ptr, w * elemsize);
-
- outptr += w;
- }
-
- return 0;
- }
-
- if (dims == 2 && axis == 0)
- {
- // concat image
- int w = bottom_blobs[0].w;
-
- // total height
- int top_h = 0;
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const Mat& bottom_blob = bottom_blobs[b];
- top_h += bottom_blob.h;
- }
-
- Mat& top_blob = top_blobs[0];
- top_blob.create(w, top_h, elemsize, opt.blob_allocator);
- if (top_blob.empty())
- return -100;
-
- float* outptr = top_blob;
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const Mat& bottom_blob = bottom_blobs[b];
-
- int size = w * bottom_blob.h;
-
- const float* ptr = bottom_blob;
- memcpy(outptr, ptr, size * elemsize);
-
- outptr += size;
- }
-
- return 0;
- }
-
- if (dims == 2 && axis == 1)
- {
- // interleave image row
- int h = bottom_blobs[0].h;
-
- // total width
- int top_w = 0;
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const Mat& bottom_blob = bottom_blobs[b];
- top_w += bottom_blob.w;
- }
-
- Mat& top_blob = top_blobs[0];
- top_blob.create(top_w, h, elemsize, opt.blob_allocator);
- if (top_blob.empty())
- return -100;
-
- #pragma omp parallel for num_threads(opt.num_threads)
- for (int i=0; i<h; i++)
- {
- float* outptr = top_blob.row(i);
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const Mat& bottom_blob = bottom_blobs[b];
-
- const float* ptr = bottom_blob.row(i);
- memcpy(outptr, ptr, bottom_blob.w * elemsize);
-
- outptr += bottom_blob.w;
- }
- }
-
- return 0;
- }
-
- if (dims == 3 && axis == 0)
- {
- // concat dim
- int w = bottom_blobs[0].w;
- int h = bottom_blobs[0].h;
-
- // total channels
- int top_channels = 0;
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const Mat& bottom_blob = bottom_blobs[b];
- top_channels += bottom_blob.c;
- }
-
- Mat& top_blob = top_blobs[0];
- top_blob.create(w, h, top_channels, elemsize, opt.blob_allocator);
- if (top_blob.empty())
- return -100;
-
- int q = 0;
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const Mat& bottom_blob = bottom_blobs[b];
-
- int channels = bottom_blob.c;
- int size = bottom_blob.cstep * channels;
-
- const float* ptr = bottom_blob;
- float* outptr = top_blob.channel(q);
- memcpy(outptr, ptr, size * elemsize);
-
- q += channels;
- }
-
- return 0;
- }
-
- if (dims == 3 && axis == 1)
- {
- // interleave dim height
- int w = bottom_blobs[0].w;
- int channels = bottom_blobs[0].c;
-
- // total height
- int top_h = 0;
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const Mat& bottom_blob = bottom_blobs[b];
- top_h += bottom_blob.h;
- }
-
- Mat& top_blob = top_blobs[0];
- top_blob.create(w, top_h, channels, elemsize, opt.blob_allocator);
- if (top_blob.empty())
- return -100;
-
- #pragma omp parallel for num_threads(opt.num_threads)
- for (int q=0; q<channels; q++)
- {
- float* outptr = top_blob.channel(q);
-
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const Mat& bottom_blob = bottom_blobs[b];
-
- int size = bottom_blob.w * bottom_blob.h;
-
- const float* ptr = bottom_blob.channel(q);
- memcpy(outptr, ptr, size * elemsize);
-
- outptr += size;
- }
- }
-
- return 0;
- }
-
- if (dims == 3 && axis == 2)
- {
- // interleave dim width
- int h = bottom_blobs[0].h;
- int channels = bottom_blobs[0].c;
-
- // total height
- int top_w = 0;
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const Mat& bottom_blob = bottom_blobs[b];
- top_w += bottom_blob.w;
- }
-
- Mat& top_blob = top_blobs[0];
- top_blob.create(top_w, h, channels, elemsize, opt.blob_allocator);
- if (top_blob.empty())
- return -100;
-
- #pragma omp parallel for num_threads(opt.num_threads)
- for (int q=0; q<channels; q++)
- {
- float* outptr = top_blob.channel(q);
-
- for (int i=0; i<h; i++)
- {
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const Mat& bottom_blob = bottom_blobs[b];
-
- const float* ptr = bottom_blob.channel(q).row(i);
- memcpy(outptr, ptr, bottom_blob.w * elemsize);
-
- outptr += bottom_blob.w;
- }
- }
- }
-
- return 0;
- }
-
- return 0;
- }
-
- #if NCNN_VULKAN
- int Concat::create_pipeline()
- {
- std::vector<vk_specialization_type> specializations(1);
- specializations[0].i = axis;
-
- // pack1
- {
- pipeline_concat = new Pipeline(vkdev);
- pipeline_concat->set_optimal_local_size_xyz();
- pipeline_concat->create("concat", specializations, 2, 11);
- }
-
- // pack4
- {
- pipeline_concat_pack4 = new Pipeline(vkdev);
- pipeline_concat_pack4->set_optimal_local_size_xyz();
- pipeline_concat_pack4->create("concat_pack4", specializations, 2, 11);
- }
-
- // pack4to1
- {
- pipeline_concat_pack4to1 = new Pipeline(vkdev);
- pipeline_concat_pack4to1->set_optimal_local_size_xyz();
- pipeline_concat_pack4to1->create("concat_pack4to1", specializations, 2, 11);
- }
-
- return 0;
- }
-
- int Concat::destroy_pipeline()
- {
- delete pipeline_concat;
- pipeline_concat = 0;
-
- delete pipeline_concat_pack4;
- pipeline_concat_pack4 = 0;
-
- delete pipeline_concat_pack4to1;
- pipeline_concat_pack4to1 = 0;
-
- return 0;
- }
-
- int Concat::forward(const std::vector<VkMat>& bottom_blobs, std::vector<VkMat>& top_blobs, VkCompute& cmd, const Option& opt) const
- {
- int dims = bottom_blobs[0].dims;
-
- if (dims == 1) // axis == 0
- {
- // concat vector
- // total length
- size_t elemsize = bottom_blobs[0].elemsize;
- int packing = bottom_blobs[0].packing;
- int top_w = 0;
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const VkMat& bottom_blob = bottom_blobs[b];
- elemsize = std::min(elemsize, bottom_blob.elemsize);
- packing = std::min(packing, bottom_blob.packing);
- top_w += bottom_blob.w * bottom_blob.packing;
- }
-
- int out_packing = top_w % 4 == 0 ? 4 : 1;
- size_t out_elemsize = elemsize / packing * out_packing;
-
- // TODO pack1to4 and pack4to1to4 make sense ?
- if (packing == 1)
- {
- out_packing = 1;
- out_elemsize = elemsize / packing;
- }
-
- VkMat& top_blob = top_blobs[0];
- top_blob.create(top_w / out_packing, out_elemsize, out_packing, opt.blob_vkallocator, opt.staging_vkallocator);
- if (top_blob.empty())
- return -100;
-
- cmd.record_prepare_compute_barrier(top_blob);
-
- int woffset = 0;
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const VkMat& bottom_blob = bottom_blobs[b];
-
- std::vector<VkMat> bindings(2);
- bindings[0] = bottom_blob;
- bindings[1] = top_blob;
-
- std::vector<vk_constant_type> constants(11);
- constants[0].i = bottom_blob.dims;
- constants[1].i = bottom_blob.w;
- constants[2].i = bottom_blob.h;
- constants[3].i = bottom_blob.c;
- constants[4].i = bottom_blob.cstep;
- constants[5].i = top_blob.dims;
- constants[6].i = top_blob.w;
- constants[7].i = top_blob.h;
- constants[8].i = top_blob.c;
- constants[9].i = top_blob.cstep;
- constants[10].i = woffset;
-
- const Pipeline* pipeline = 0;
- if (bottom_blob.packing == 1 && out_packing == 1)
- {
- pipeline = pipeline_concat;
- }
- else if (bottom_blob.packing == 4 && out_packing == 4)
- {
- pipeline = pipeline_concat_pack4;
- }
- else if (bottom_blob.packing == 4 && out_packing == 1)
- {
- pipeline = pipeline_concat_pack4to1;
- }
-
- // record
- cmd.record_prepare_compute_barrier(bottom_blob);
- cmd.record_pipeline(pipeline, bindings, constants, bottom_blob);
-
- woffset += bottom_blob.w * bottom_blob.packing / out_packing;
- }
-
- return 0;
- }
-
- if (dims == 2 && axis == 0)
- {
- // concat image
- int w = bottom_blobs[0].w;
-
- // total height
- size_t elemsize = bottom_blobs[0].elemsize;
- int packing = bottom_blobs[0].packing;
- int top_h = 0;
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const VkMat& bottom_blob = bottom_blobs[b];
- elemsize = std::min(elemsize, bottom_blob.elemsize);
- packing = std::min(packing, bottom_blob.packing);
- top_h += bottom_blob.h * bottom_blob.packing;
- }
-
- int out_packing = top_h % 4 == 0 ? 4 : 1;
- size_t out_elemsize = elemsize / packing * out_packing;
-
- // TODO pack1to4 and pack4to1to4 make sense ?
- if (packing == 1)
- {
- out_packing = 1;
- out_elemsize = elemsize / packing;
- }
-
- VkMat& top_blob = top_blobs[0];
- top_blob.create(w, top_h / out_packing, out_elemsize, out_packing, opt.blob_vkallocator, opt.staging_vkallocator);
- if (top_blob.empty())
- return -100;
-
- cmd.record_prepare_compute_barrier(top_blob);
-
- int hoffset = 0;
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const VkMat& bottom_blob = bottom_blobs[b];
-
- std::vector<VkMat> bindings(2);
- bindings[0] = bottom_blob;
- bindings[1] = top_blob;
-
- std::vector<vk_constant_type> constants(11);
- constants[0].i = bottom_blob.dims;
- constants[1].i = bottom_blob.w;
- constants[2].i = bottom_blob.h;
- constants[3].i = bottom_blob.c;
- constants[4].i = bottom_blob.cstep;
- constants[5].i = top_blob.dims;
- constants[6].i = top_blob.w;
- constants[7].i = top_blob.h;
- constants[8].i = top_blob.c;
- constants[9].i = top_blob.cstep;
- constants[10].i = hoffset;
-
- const Pipeline* pipeline = 0;
- if (bottom_blob.packing == 1 && out_packing == 1)
- {
- pipeline = pipeline_concat;
- }
- else if (bottom_blob.packing == 4 && out_packing == 4)
- {
- pipeline = pipeline_concat_pack4;
- }
- else if (bottom_blob.packing == 4 && out_packing == 1)
- {
- pipeline = pipeline_concat_pack4to1;
- }
-
- // record
- cmd.record_prepare_compute_barrier(bottom_blob);
- cmd.record_pipeline(pipeline, bindings, constants, bottom_blob);
-
- hoffset += bottom_blob.h * bottom_blob.packing / out_packing;
- }
-
- return 0;
- }
-
- if (dims == 2 && axis == 1)
- {
- // interleave image row
- int h = bottom_blobs[0].h;
- size_t elemsize = bottom_blobs[0].elemsize;
- int packing = bottom_blobs[0].packing;
-
- // total width
- int top_w = 0;
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const VkMat& bottom_blob = bottom_blobs[b];
- top_w += bottom_blob.w;
- }
-
- VkMat& top_blob = top_blobs[0];
- top_blob.create(top_w, h, elemsize, packing, opt.blob_vkallocator, opt.staging_vkallocator);
- if (top_blob.empty())
- return -100;
-
- cmd.record_prepare_compute_barrier(top_blob);
-
- int woffset = 0;
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const VkMat& bottom_blob = bottom_blobs[b];
-
- std::vector<VkMat> bindings(2);
- bindings[0] = bottom_blob;
- bindings[1] = top_blob;
-
- std::vector<vk_constant_type> constants(11);
- constants[0].i = bottom_blob.dims;
- constants[1].i = bottom_blob.w;
- constants[2].i = bottom_blob.h;
- constants[3].i = bottom_blob.c;
- constants[4].i = bottom_blob.cstep;
- constants[5].i = top_blob.dims;
- constants[6].i = top_blob.w;
- constants[7].i = top_blob.h;
- constants[8].i = top_blob.c;
- constants[9].i = top_blob.cstep;
- constants[10].i = woffset;
-
- const Pipeline* pipeline = packing == 4 ? pipeline_concat_pack4 : pipeline_concat;
-
- // record
- cmd.record_prepare_compute_barrier(bottom_blob);
- cmd.record_pipeline(pipeline, bindings, constants, bottom_blob);
-
- woffset += bottom_blob.w;
- }
-
- return 0;
- }
-
- if (dims == 3 && axis == 0)
- {
- // concat dim
- int w = bottom_blobs[0].w;
- int h = bottom_blobs[0].h;
-
- // total channels
- size_t elemsize = bottom_blobs[0].elemsize;
- int packing = bottom_blobs[0].packing;
- int top_channels = 0;
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const VkMat& bottom_blob = bottom_blobs[b];
- elemsize = std::min(elemsize, bottom_blob.elemsize);
- packing = std::min(packing, bottom_blob.packing);
- top_channels += bottom_blob.c * bottom_blob.packing;
- }
-
- int out_packing = top_channels % 4 == 0 ? 4 : 1;
- size_t out_elemsize = elemsize / packing * out_packing;
-
- // TODO pack1to4 and pack4to1to4 make sense ?
- if (packing == 1)
- {
- out_packing = 1;
- out_elemsize = elemsize / packing;
- }
-
- VkMat& top_blob = top_blobs[0];
- top_blob.create(w, h, top_channels / out_packing, out_elemsize, out_packing, opt.blob_vkallocator, opt.staging_vkallocator);
- if (top_blob.empty())
- return -100;
-
- cmd.record_prepare_compute_barrier(top_blob);
-
- int coffset = 0;
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const VkMat& bottom_blob = bottom_blobs[b];
-
- std::vector<VkMat> bindings(2);
- bindings[0] = bottom_blob;
- bindings[1] = top_blob;
-
- std::vector<vk_constant_type> constants(11);
- constants[0].i = bottom_blob.dims;
- constants[1].i = bottom_blob.w;
- constants[2].i = bottom_blob.h;
- constants[3].i = bottom_blob.c;
- constants[4].i = bottom_blob.cstep;
- constants[5].i = top_blob.dims;
- constants[6].i = top_blob.w;
- constants[7].i = top_blob.h;
- constants[8].i = top_blob.c;
- constants[9].i = top_blob.cstep;
- constants[10].i = coffset;
-
- const Pipeline* pipeline = 0;
- if (bottom_blob.packing == 1 && out_packing == 1)
- {
- pipeline = pipeline_concat;
- }
- else if (bottom_blob.packing == 4 && out_packing == 4)
- {
- pipeline = pipeline_concat_pack4;
- }
- else if (bottom_blob.packing == 4 && out_packing == 1)
- {
- pipeline = pipeline_concat_pack4to1;
- }
-
- // record
- cmd.record_prepare_compute_barrier(bottom_blob);
- cmd.record_pipeline(pipeline, bindings, constants, bottom_blob);
-
- coffset += bottom_blob.c * bottom_blob.packing / out_packing;
- }
-
- return 0;
- }
-
- if (dims == 3 && axis == 1)
- {
- // interleave dim height
- int w = bottom_blobs[0].w;
- int channels = bottom_blobs[0].c;
- size_t elemsize = bottom_blobs[0].elemsize;
- int packing = bottom_blobs[0].packing;
-
- // total height
- int top_h = 0;
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const VkMat& bottom_blob = bottom_blobs[b];
- top_h += bottom_blob.h;
- }
-
- VkMat& top_blob = top_blobs[0];
- top_blob.create(w, top_h, channels, elemsize, packing, opt.blob_vkallocator, opt.staging_vkallocator);
- if (top_blob.empty())
- return -100;
-
- cmd.record_prepare_compute_barrier(top_blob);
-
- int hoffset = 0;
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const VkMat& bottom_blob = bottom_blobs[b];
-
- std::vector<VkMat> bindings(2);
- bindings[0] = bottom_blob;
- bindings[1] = top_blob;
-
- std::vector<vk_constant_type> constants(11);
- constants[0].i = bottom_blob.dims;
- constants[1].i = bottom_blob.w;
- constants[2].i = bottom_blob.h;
- constants[3].i = bottom_blob.c;
- constants[4].i = bottom_blob.cstep;
- constants[5].i = top_blob.dims;
- constants[6].i = top_blob.w;
- constants[7].i = top_blob.h;
- constants[8].i = top_blob.c;
- constants[9].i = top_blob.cstep;
- constants[10].i = hoffset;
-
- const Pipeline* pipeline = packing == 4 ? pipeline_concat_pack4 : pipeline_concat;
-
- // record
- cmd.record_prepare_compute_barrier(bottom_blob);
- cmd.record_pipeline(pipeline, bindings, constants, bottom_blob);
-
- hoffset += bottom_blob.h;
- }
-
- return 0;
- }
-
- if (dims == 3 && axis == 2)
- {
- // interleave dim width
- int h = bottom_blobs[0].h;
- int channels = bottom_blobs[0].c;
- size_t elemsize = bottom_blobs[0].elemsize;
- int packing = bottom_blobs[0].packing;
-
- // total height
- int top_w = 0;
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const VkMat& bottom_blob = bottom_blobs[b];
- top_w += bottom_blob.w;
- }
-
- VkMat& top_blob = top_blobs[0];
- top_blob.create(top_w, h, channels, elemsize, packing, opt.blob_vkallocator, opt.staging_vkallocator);
- if (top_blob.empty())
- return -100;
-
- cmd.record_prepare_compute_barrier(top_blob);
-
- int woffset = 0;
- for (size_t b=0; b<bottom_blobs.size(); b++)
- {
- const VkMat& bottom_blob = bottom_blobs[b];
-
- std::vector<VkMat> bindings(2);
- bindings[0] = bottom_blob;
- bindings[1] = top_blob;
-
- std::vector<vk_constant_type> constants(11);
- constants[0].i = bottom_blob.dims;
- constants[1].i = bottom_blob.w;
- constants[2].i = bottom_blob.h;
- constants[3].i = bottom_blob.c;
- constants[4].i = bottom_blob.cstep;
- constants[5].i = top_blob.dims;
- constants[6].i = top_blob.w;
- constants[7].i = top_blob.h;
- constants[8].i = top_blob.c;
- constants[9].i = top_blob.cstep;
- constants[10].i = woffset;
-
- const Pipeline* pipeline = packing == 4 ? pipeline_concat_pack4 : pipeline_concat;
-
- // record
- cmd.record_prepare_compute_barrier(bottom_blob);
- cmd.record_pipeline(pipeline, bindings, constants, bottom_blob);
-
- woffset += bottom_blob.w;
- }
-
- return 0;
- }
-
- return 0;
- }
- #endif // NCNN_VULKAN
-
- } // namespace ncnn
|