tensor_stream-opencl 0.1.1 → 0.1.2

checksums.yaml

@@ -1,7 +1,7 @@
 ---
-SHA1:
-  metadata.gz: d33b253833ef9d0db6be85834b2af716d5c2b1fb
-  data.tar.gz: b2e18fcf323ff2353bd251c0d6a9d01c6ef7d729
+SHA256:
+  metadata.gz: 466d846646b2b58659b7068260e84b1a76b64b053c0c8c2dc9c60d7b817041c1
+  data.tar.gz: 2a440d5cd6d809dbd661576e38277edb9a738b453d21c27f3048bfcaca9f4852
 SHA512:
-  metadata.gz: 3933ab4454e758dd9bdfd0bf3bff4c856369da6704f6ded41e35ff50a271b259381eadda0e285a95d92a4f277975da56fb084cc03fb4b408403ace6021a146d9
-  data.tar.gz: 5f39e60e3b435d24a49cbeb0af399545c98a031999a26ff07f4c4d6badef58ce3799b5c39dfbe2f3899a2dd44b729b61cdd7e6b9a47f8a5d35e5b6487626fcc4
+  metadata.gz: d25b928b3b3b9ca3658f14d42e898a067b356eb0f806356e5a093687d652d47b151c34c73394d0bd33048cf6dc68e918e6427287f07d5e90d9741b056460ab2f
+  data.tar.gz: 5fda92ffbc2c3063e6bcedf8ce00c024bab356425f9d723988d755e43e62b7ff283c2087e1ec455f3680e718d8256b914220b2a8e3bcd60edce7287225ec1a61

data/lib/tensor_stream/opencl/kernels/apply_adadelta.cl

@@ -0,0 +1,21 @@
+% c_dtype = dtype_to_c_type(dtype)
+ // same dimension add floating point op
+ __kernel void apply_adadelta_<%= dtype %>(const int M, const int N,
+                                       __global const <%= c_dtype %> *lr,
+                                       __global const <%= c_dtype %> *rho,
+                                       __global const <%= c_dtype %> *epsilon,
+                                       __global const <%= c_dtype %> *grad,
+                                       __global <%= c_dtype %> *output,
+                                       __global <%= c_dtype %> *acc,
+                                       __global <%= c_dtype %> *acc_update
+                                       ) {
+    // Get the index of the current element to be processed
+    const int globalRow = get_global_id(0); // Row ID of C (0..M)
+    const int globalCol = get_global_id(1); // Col ID of C (0..N)
+    const int index = globalRow * N + globalCol;
+
+    acc[index] = acc[index] * rho[0] + (grad[index] * grad[index]) * ((<%= c_dtype %>)1 - rho[0]);
+    const <%= c_dtype %> update = sqrt(acc_update[index] + epsilon[0]) * rsqrt(acc[index] + epsilon[0]) * grad[index];
+    output[index] -= update * lr[0];
+    acc_update[index] = acc_update[index] * rho[0] + update * update * ((<%= c_dtype %>)1 - rho[0]);
+}

data/lib/tensor_stream/opencl/nn_ops.rb

@@ -1,3 +1,4 @@
+require 'pry-byebug'
 module TensorStream
   module OpenCLHelpers
     # Collection of math functions for interfacing with OpenCL kernels
@@ -54,6 +55,39 @@ module TensorStream
           end
 
           register_op :apply_adadelta do |context, tensor, inputs|
+            _target_var, _accum, _accum_update, lr, rho, epsilon, grad = inputs
+            assign = tensor.inputs[0] || tensor
+            assign_acc = tensor.inputs[1]
+            assign_acc_update = tensor.inputs[2]
+
+            # mark variable buffers as dirty
+            assign.buffer.dirty = true # force buffer copy when variable is read externally
+            assign_acc.buffer.dirty = true # force buffer copy when variable is read externally
+            assign_acc_update.buffer.dirty = true # force buffer copy when variable is read externally
+
+            output_buffer = assign.buffer
+
+            m, n = output_buffer.shape
+            work_group = [m || 1, n || 1]
+            cl_m = OpenCL::Int1.new(m || 1)
+            cl_n = OpenCL::Int1.new(n || 1)
+
+            event_wait_list = build_event_wait_list(inputs)
+            method_call = :"apply_adadelta_#{output_buffer.data_type}"
+            event = _cl_program('apply_adadelta', dtype: output_buffer.data_type)
+                                .send(method_call, _opencl_queue, work_group, cl_m, cl_n,
+                                      lr.cl_buffer,
+                                      rho.cl_buffer,
+                                      epsilon.cl_buffer,
+                                      grad.cl_buffer,
+                                      assign.buffer.cl_buffer,
+                                      assign_acc.buffer.cl_buffer,
+                                      assign_acc_update.buffer.cl_buffer,
+                                      event_wait_list: event_wait_list)
+            output_buffer.op = event
+            assign_acc.buffer.op = event
+            assign_acc_update.buffer.op = event
+            output_buffer
           end
 
           # Adam optimization algorithm

data/lib/tensor_stream/opencl/version.rb

@@ -1,5 +1,5 @@
 module TensorStream
   module Opencl
-    VERSION = "0.1.1"
+    VERSION = "0.1.2"
   end
 end

data/tensor_stream-opencl.gemspec

@@ -35,6 +35,7 @@ Gem::Specification.new do |spec|
   spec.add_development_dependency "bundler", "~> 1.16"
   spec.add_development_dependency "rake", "~> 10.0"
   spec.add_development_dependency "rspec", "~> 3.0"
+  spec.add_development_dependency "pry-byebug"
   spec.add_dependency "tensor_stream"
   spec.add_dependency "opencl_ruby_ffi"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: tensor_stream-opencl
 version: !ruby/object:Gem::Version
-  version: 0.1.1
+  version: 0.1.2
 platform: ruby
 authors:
 - Joseph Dayo
@@ -52,6 +52,20 @@ dependencies:
     - - "~>"
       - !ruby/object:Gem::Version
         version: '3.0'
+- !ruby/object:Gem::Dependency
+  name: pry-byebug
+  requirement: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
+  type: :development
+  prerelease: false
+  version_requirements: !ruby/object:Gem::Requirement
+    requirements:
+    - - ">="
+      - !ruby/object:Gem::Version
+        version: '0'
 - !ruby/object:Gem::Dependency
   name: tensor_stream
   requirement: !ruby/object:Gem::Requirement
@@ -104,6 +118,7 @@ files:
 - lib/tensor_stream/opencl/kernels/abs.cl
 - lib/tensor_stream/opencl/kernels/acos.cl
 - lib/tensor_stream/opencl/kernels/add.cl
+- lib/tensor_stream/opencl/kernels/apply_adadelta.cl
 - lib/tensor_stream/opencl/kernels/apply_adam.cl
 - lib/tensor_stream/opencl/kernels/apply_gradient.cl
 - lib/tensor_stream/opencl/kernels/apply_momentum.cl
@@ -178,7 +193,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
       version: '0'
 requirements: []
 rubyforge_project: 
-rubygems_version: 2.6.10
+rubygems_version: 2.7.7
 signing_key: 
 specification_version: 4
 summary: OpenCL evaluator for tensor_stream