diff --git a/DRED_codegen/.cproject b/.cproject
similarity index 100%
rename from DRED_codegen/.cproject
rename to .cproject
diff --git a/.idea/workspace.xml b/.idea/workspace.xml
index f2f1e7de626a51c4a6b0e9df5ca421b83ce35a48..3153075588b5f6041e13da35ed568fe49126a46a 100644
--- a/.idea/workspace.xml
+++ b/.idea/workspace.xml
@@ -3,51 +3,6 @@
<component name="ChangeListManager">
<list default="true" id="b045d9d1-ede8-4e19-b09a-ddf511e7d34c" name="Default Changelist" comment="">
<change beforePath="$PROJECT_DIR$/.idea/workspace.xml" beforeDir="false" afterPath="$PROJECT_DIR$/.idea/workspace.xml" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/history_trace.npy" beforeDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/gened.cpp" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/gened.cpp" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.export.ll" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.export.ll" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.0.bc" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.0.bc" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.1.bc" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.1.bc" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.1.ll" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.1.ll" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.2.bc" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.2.bc" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.2.prechk.bc" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.2.prechk.bc" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.bc" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.bc" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.ll" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.ll" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.1.bc" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.1.bc" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.1.ll" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.1.ll" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.1.tmp.bc" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.1.tmp.bc" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.2.bc" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.2.bc" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.2.ll" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.2.ll" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.3.bc" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.3.bc" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.3.ll" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.3.ll" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.bc" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.bc" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.ll" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.ll" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.pp.0.bc" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.pp.0.bc" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.pp.bc" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/a.pp.bc" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/autopilot.flow.log" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/autopilot.flow.log" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/autopilot.rtl.models.txt" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/autopilot.rtl.models.txt" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/bugpoint.sh" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/bugpoint.sh" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.bc" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.bc" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.g.bc" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.g.bc" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pp.0.cpp" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pp.0.cpp" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pp.0.cpp.ap-cdt.cpp" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pp.0.cpp.ap-cdt.cpp" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pp.0.cpp.ap-line.cpp" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pp.0.cpp.ap-line.cpp" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pp.00.o" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pp.00.o" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pragma.0.cpp" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pragma.0.cpp" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pragma.1.cpp" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pragma.1.cpp" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pragma.2.cpp" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pragma.2.cpp" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/global.setting.tcl" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/global.setting.tcl" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/hls_design_meta.cpp" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/hls_design_meta.cpp" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/hls_design_meta.h" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/hls_design_meta.h" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/hls_design_meta.tcl" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/hls_design_meta.tcl" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/pragma.0.directive" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/pragma.0.directive" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/pragma.1.directive" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/.autopilot/db/pragma.1.directive" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/solution1.aps" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/solution1.aps" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/solution1.log" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/hls_proj/solution1/solution1.log" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/inference.py" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/inference.py" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/script.tcl" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/script.tcl" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/synth.log" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/synth.log" afterDir="false" />
- <change beforePath="$PROJECT_DIR$/DRED_codegen/vivado_hls.log" beforeDir="false" afterPath="$PROJECT_DIR$/DRED_codegen/vivado_hls.log" afterDir="false" />
</list>
<option name="EXCLUDED_CONVERTED_TO_IGNORED" value="true" />
<option name="SHOW_DIALOG" value="false" />
@@ -95,7 +50,7 @@
<session id="1950885685">
<usages-collector id="statistics.lifecycle.project">
<counts>
- <entry key="project.closed" value="10" />
+ <entry key="project.closed" value="11" />
<entry key="project.open.time.0" value="1" />
<entry key="project.open.time.15" value="1" />
<entry key="project.open.time.17" value="1" />
@@ -143,13 +98,13 @@
<entry key="py" value="4531" />
<entry key="tcl" value="30" />
<entry key="test logs" value="17" />
- <entry key="txt" value="482" />
+ <entry key="txt" value="529" />
</counts>
</usages-collector>
<usages-collector id="statistics.file.types.edit">
<counts>
<entry key="C++" value="2" />
- <entry key="PLAIN_TEXT" value="643" />
+ <entry key="PLAIN_TEXT" value="690" />
<entry key="Python" value="4667" />
</counts>
</usages-collector>
@@ -198,8 +153,8 @@
<file pinned="false" current-in-tab="true">
<entry file="file://$PROJECT_DIR$/DRED_codegen/exhaustive_search.py">
<provider selected="true" editor-type-id="text-editor">
- <state relative-caret-position="513">
- <caret line="141" column="19" selection-start-line="141" selection-start-column="19" selection-end-line="141" selection-end-column="19" />
+ <state relative-caret-position="548">
+ <caret line="160" column="30" lean-forward="true" selection-start-line="160" selection-start-column="30" selection-end-line="160" selection-end-column="30" />
<folding>
<element signature="e#1#59#0" expanded="true" />
</folding>
@@ -382,7 +337,7 @@
<option name="IS_MODULE_SDK" value="true" />
<option name="ADD_CONTENT_ROOTS" value="true" />
<option name="ADD_SOURCE_ROOTS" value="true" />
- <option name="SCRIPT_NAME" value="$PROJECT_DIR$/DRED_codegen/codegen.py" />
+ <option name="SCRIPT_NAME" value="$PROJECT_DIR$/codegen.py" />
<option name="PARAMETERS" value="" />
<option name="SHOW_COMMAND_LINE" value="false" />
<option name="EMULATE_TERMINAL" value="false" />
@@ -403,7 +358,7 @@
<option name="IS_MODULE_SDK" value="true" />
<option name="ADD_CONTENT_ROOTS" value="true" />
<option name="ADD_SOURCE_ROOTS" value="true" />
- <option name="SCRIPT_NAME" value="$PROJECT_DIR$/DRED_codegen/exhaustive_search.py" />
+ <option name="SCRIPT_NAME" value="$PROJECT_DIR$/exhaustive_search.py" />
<option name="PARAMETERS" value="" />
<option name="SHOW_COMMAND_LINE" value="false" />
<option name="EMULATE_TERMINAL" value="false" />
@@ -425,7 +380,7 @@
<option name="IS_MODULE_SDK" value="true" />
<option name="ADD_CONTENT_ROOTS" value="true" />
<option name="ADD_SOURCE_ROOTS" value="true" />
- <option name="SCRIPT_NAME" value="$PROJECT_DIR$/DRED_codegen/inference.py" />
+ <option name="SCRIPT_NAME" value="$PROJECT_DIR$/inference.py" />
<option name="PARAMETERS" value="" />
<option name="SHOW_COMMAND_LINE" value="false" />
<option name="EMULATE_TERMINAL" value="false" />
@@ -447,7 +402,7 @@
<option name="IS_MODULE_SDK" value="true" />
<option name="ADD_CONTENT_ROOTS" value="true" />
<option name="ADD_SOURCE_ROOTS" value="true" />
- <option name="SCRIPT_NAME" value="$PROJECT_DIR$/DRED_codegen/q_learning.py" />
+ <option name="SCRIPT_NAME" value="$PROJECT_DIR$/q_learning.py" />
<option name="PARAMETERS" value="" />
<option name="SHOW_COMMAND_LINE" value="false" />
<option name="EMULATE_TERMINAL" value="false" />
@@ -469,7 +424,7 @@
<option name="IS_MODULE_SDK" value="true" />
<option name="ADD_CONTENT_ROOTS" value="true" />
<option name="ADD_SOURCE_ROOTS" value="true" />
- <option name="SCRIPT_NAME" value="$PROJECT_DIR$/DRED_codegen/q_learning_v2.py" />
+ <option name="SCRIPT_NAME" value="$PROJECT_DIR$/q_learning_v2.py" />
<option name="PARAMETERS" value="" />
<option name="SHOW_COMMAND_LINE" value="false" />
<option name="EMULATE_TERMINAL" value="false" />
@@ -490,7 +445,7 @@
<option name="IS_MODULE_SDK" value="true" />
<option name="ADD_CONTENT_ROOTS" value="true" />
<option name="ADD_SOURCE_ROOTS" value="true" />
- <option name="SCRIPT_NAME" value="$PROJECT_DIR$/DRED_codegen/tempate.py" />
+ <option name="SCRIPT_NAME" value="$PROJECT_DIR$/tempate.py" />
<option name="PARAMETERS" value="" />
<option name="SHOW_COMMAND_LINE" value="false" />
<option name="EMULATE_TERMINAL" value="false" />
@@ -599,7 +554,14 @@
<option name="project" value="LOCAL" />
<updated>1544975212912</updated>
</task>
- <option name="localTasksCounter" value="11" />
+ <task id="LOCAL-00011" summary="Final commit before submitting the report">
+ <created>1545337067351</created>
+ <option name="number" value="00011" />
+ <option name="presentableId" value="LOCAL-00011" />
+ <option name="project" value="LOCAL" />
+ <updated>1545337067351</updated>
+ </task>
+ <option name="localTasksCounter" value="12" />
<servers />
</component>
<component name="TodoView">
@@ -613,13 +575,14 @@
</component>
<component name="ToolWindowManager">
<frame x="121" y="13" width="3720" height="2148" extended-state="6" />
+ <editor active="true" />
<layout>
<window_info active="true" content_ui="combo" id="Project" order="0" visible="true" weight="0.2044077" />
<window_info id="Structure" order="1" side_tool="true" weight="0.25" />
<window_info id="Favorites" order="2" side_tool="true" />
<window_info anchor="bottom" id="Message" order="0" />
<window_info anchor="bottom" id="Find" order="1" weight="0.3296875" />
- <window_info anchor="bottom" id="Run" order="2" visible="true" weight="0.3421875" />
+ <window_info anchor="bottom" id="Run" order="2" weight="0.3421875" />
<window_info anchor="bottom" id="Debug" order="3" weight="0.4" />
<window_info anchor="bottom" id="Cvs" order="4" weight="0.25" />
<window_info anchor="bottom" id="Inspection" order="5" weight="0.4" />
@@ -686,7 +649,8 @@
<MESSAGE value="bugs in the newly added if logic" />
<MESSAGE value="fixed history lookup feature" />
<MESSAGE value="finished training for now, ends at ~8380, traces and logs added" />
- <option name="LAST_COMMIT_MESSAGE" value="finished training for now, ends at ~8380, traces and logs added" />
+ <MESSAGE value="Final commit before submitting the report" />
+ <option name="LAST_COMMIT_MESSAGE" value="Final commit before submitting the report" />
</component>
<component name="editorHistoryManager">
<entry file="file://$PROJECT_DIR$/Board Coverage" />
@@ -831,8 +795,8 @@
</entry>
<entry file="file://$PROJECT_DIR$/DRED_codegen/exhaustive_search.py">
<provider selected="true" editor-type-id="text-editor">
- <state relative-caret-position="513">
- <caret line="141" column="19" selection-start-line="141" selection-start-column="19" selection-end-line="141" selection-end-column="19" />
+ <state relative-caret-position="548">
+ <caret line="160" column="30" lean-forward="true" selection-start-line="160" selection-start-column="30" selection-end-line="160" selection-end-column="30" />
<folding>
<element signature="e#1#59#0" expanded="true" />
</folding>
diff --git a/DRED_codegen/.project b/.project
similarity index 100%
rename from DRED_codegen/.project
rename to .project
diff --git a/DRED_codegen/1k-3k_trace.log b/1k-3k_trace.log
similarity index 100%
rename from DRED_codegen/1k-3k_trace.log
rename to 1k-3k_trace.log
diff --git a/DRED_codegen/5k_7k_trace.log b/5k_7k_trace.log
similarity index 100%
rename from DRED_codegen/5k_7k_trace.log
rename to 5k_7k_trace.log
diff --git a/DRED_codegen/8k.log b/8k.log
similarity index 100%
rename from DRED_codegen/8k.log
rename to 8k.log
diff --git a/DRED_codegen/.apc/autopilot.apfmapping b/DRED_codegen/.apc/autopilot.apfmapping
deleted file mode 100644
index 067af1c0b4890d04b3afda69e30a49543d0fcd95..0000000000000000000000000000000000000000
--- a/DRED_codegen/.apc/autopilot.apfmapping
+++ /dev/null
@@ -1,7 +0,0 @@
-<?xml version="1.0" encoding="UTF-8"?>
-<autopilotfilemapping:AutoPilotFileMapping xmlns:autopilotfilemapping="www.autoesl.com/autopilotfilemapping">
- <source>
- <originFiles name="gened.cpp" path="G:\ECE527\DRED\DRED_codegen\gened.cpp"/>
- </source>
- <testbench/>
-</autopilotfilemapping:AutoPilotFileMapping>
diff --git a/DRED_codegen/.idea/modules.xml b/DRED_codegen/.idea/modules.xml
deleted file mode 100644
index 7b6e3e2b610783953af28ff9b3f10ea67972b6c3..0000000000000000000000000000000000000000
--- a/DRED_codegen/.idea/modules.xml
+++ /dev/null
@@ -1,8 +0,0 @@
-<?xml version="1.0" encoding="UTF-8"?>
-<project version="4">
- <component name="ProjectModuleManager">
- <modules>
- <module fileurl="file://$PROJECT_DIR$/.idea/DRED_codegen.iml" filepath="$PROJECT_DIR$/.idea/DRED_codegen.iml" />
- </modules>
- </component>
-</project>
\ No newline at end of file
diff --git a/DRED_codegen/.idea/vcs.xml b/DRED_codegen/.idea/vcs.xml
deleted file mode 100644
index 6c0b8635858dc7ad44b93df54b762707ce49eefc..0000000000000000000000000000000000000000
--- a/DRED_codegen/.idea/vcs.xml
+++ /dev/null
@@ -1,6 +0,0 @@
-<?xml version="1.0" encoding="UTF-8"?>
-<project version="4">
- <component name="VcsDirectoryMappings">
- <mapping directory="$PROJECT_DIR$/.." vcs="Git" />
- </component>
-</project>
\ No newline at end of file
diff --git a/DRED_codegen/.settings/DRED.Debug.launch b/DRED_codegen/.settings/DRED.Debug.launch
deleted file mode 100644
index 2c0e8b7de810c00076775964915b1a52df91dc03..0000000000000000000000000000000000000000
--- a/DRED_codegen/.settings/DRED.Debug.launch
+++ /dev/null
@@ -1,12 +0,0 @@
-<?xml version="1.0" encoding="UTF-8" standalone="no"?>
-<launchConfiguration type="org.eclipse.cdt.launch.applicationLaunchType">
-<stringAttribute key="org.eclipse.cdt.launch.COREFILE_PATH" value=""/>
-<stringAttribute key="org.eclipse.cdt.launch.PROGRAM_NAME" value="Debug/a.exe"/>
-<stringAttribute key="org.eclipse.cdt.launch.PROJECT_ATTR" value="DRED"/>
-<booleanAttribute key="org.eclipse.cdt.launch.PROJECT_BUILD_CONFIG_AUTO_ATTR" value="false"/>
-<stringAttribute key="org.eclipse.cdt.launch.PROJECT_BUILD_CONFIG_ID_ATTR" value="cdt.managedbuild.config.gnu.exe.debug.1188291541"/>
-<stringAttribute key="org.eclipse.cdt.launch.WORKING_DIRECTORY" value="G:\ECE527\DRED"/>
-<booleanAttribute key="org.eclipse.cdt.launch.use_terminal" value="true"/>
-<stringAttribute key="org.eclipse.debug.core.source_locator_id" value="org.eclipse.cdt.debug.core.sourceLocator"/>
-<stringAttribute key="org.eclipse.debug.core.source_locator_memento" value="<?xml version="1.0" encoding="UTF-8" standalone="no"?> <sourceLookupDirector> <sourceContainers duplicates="false"> <container memento="&lt;?xml version=&quot;1.0&quot; encoding=&quot;UTF-8&quot; standalone=&quot;no&quot;?&gt;&#13;&#10;&lt;default/&gt;&#13;&#10;" typeId="org.eclipse.debug.core.containerType.default"/> <container memento="&lt;?xml version=&quot;1.0&quot; encoding=&quot;UTF-8&quot; standalone=&quot;no&quot;?&gt;&#13;&#10;&lt;directory nest=&quot;true&quot; path=&quot;D:\Vivado_HLS\2017.2\win64\tools\systemc&quot;/&gt;&#13;&#10;" typeId="org.eclipse.debug.core.containerType.directory"/> </sourceContainers> </sourceLookupDirector> "/>
-</launchConfiguration>
diff --git a/DRED_codegen/.settings/DRED.Release.launch b/DRED_codegen/.settings/DRED.Release.launch
deleted file mode 100644
index 451fc12a62ca5c17ea2a9ea9e0c0d3a83cd86327..0000000000000000000000000000000000000000
--- a/DRED_codegen/.settings/DRED.Release.launch
+++ /dev/null
@@ -1,12 +0,0 @@
-<?xml version="1.0" encoding="UTF-8" standalone="no"?>
-<launchConfiguration type="org.eclipse.cdt.launch.applicationLaunchType">
-<stringAttribute key="org.eclipse.cdt.launch.COREFILE_PATH" value=""/>
-<stringAttribute key="org.eclipse.cdt.launch.PROGRAM_NAME" value="Release/a.exe"/>
-<stringAttribute key="org.eclipse.cdt.launch.PROJECT_ATTR" value="DRED"/>
-<booleanAttribute key="org.eclipse.cdt.launch.PROJECT_BUILD_CONFIG_AUTO_ATTR" value="false"/>
-<stringAttribute key="org.eclipse.cdt.launch.PROJECT_BUILD_CONFIG_ID_ATTR" value="cdt.managedbuild.config.gnu.exe.release.200798804"/>
-<stringAttribute key="org.eclipse.cdt.launch.WORKING_DIRECTORY" value="G:\ECE527\DRED"/>
-<booleanAttribute key="org.eclipse.cdt.launch.use_terminal" value="true"/>
-<stringAttribute key="org.eclipse.debug.core.source_locator_id" value="org.eclipse.cdt.debug.core.sourceLocator"/>
-<stringAttribute key="org.eclipse.debug.core.source_locator_memento" value="<?xml version="1.0" encoding="UTF-8" standalone="no"?> <sourceLookupDirector> <sourceContainers duplicates="false"> <container memento="&lt;?xml version=&quot;1.0&quot; encoding=&quot;UTF-8&quot; standalone=&quot;no&quot;?&gt;&#13;&#10;&lt;default/&gt;&#13;&#10;" typeId="org.eclipse.debug.core.containerType.default"/> <container memento="&lt;?xml version=&quot;1.0&quot; encoding=&quot;UTF-8&quot; standalone=&quot;no&quot;?&gt;&#13;&#10;&lt;directory nest=&quot;true&quot; path=&quot;D:\Vivado_HLS\2017.2\win64\tools\systemc&quot;/&gt;&#13;&#10;" typeId="org.eclipse.debug.core.containerType.directory"/> </sourceContainers> </sourceLookupDirector> "/>
-</launchConfiguration>
diff --git a/DRED_codegen/main.py b/DRED_codegen/main.py
deleted file mode 100644
index 1a2a34f16b5bf39691129897161fc702dc4aa32d..0000000000000000000000000000000000000000
--- a/DRED_codegen/main.py
+++ /dev/null
@@ -1,224 +0,0 @@
-#!/usr/bin/env python
-
-"""
-Inspirit DNN Generator
-"""
-
-import math, os, platform, random, re, sys, time, threading, itertools, json, argparse, collections, glob, ast
-
-from jinja2 import Environment, FileSystemLoader, Template
-
-def generate_string_from_templatefile(template_file, target_name, render_dict = None):
- '''
- Generates a string from a template file.
- '''
- cwd = os.getcwd()
- template_dir = os.path.join(cwd, "dnngen", "templates", target_name)
- config_template_env = Environment(loader = FileSystemLoader(template_dir), trim_blocks=True, keep_trailing_newline=True)
-
- config_template_env.filters['joinpath'] = lambda list: os.path.join(*list)
- grab_template = config_template_env.get_template(template_file)
- return grab_template.render(render_dict)
-
-def generate_script_from_templatefile(template_file, output_script_path, target_name, render_dict = None) :
- '''
- Generates a test script from a template file.
- '''
-
- rendered_script = generate_string_from_templatefile(template_file, target_name, render_dict)
- with open(output_script_path, 'w') as generate_script_file:
- generate_script_file.write(rendered_script)
-
- return rendered_script
-
-def scan_for_templatefile(base_dir, templated_dir_path, target_name, render_dict = None) :
- '''
- In general, it will use the name of the tool and step to scan through
- the specific folder for templates that need to be rendered.
- '''
- files = []
- fileslist = []
- files = glob.glob(os.path.join(base_dir, templated_dir_path, target_name, "*.jinja"))
- for each in files:
- fileslist.append(os.path.basename(each))
-
- print "In " + base_dir + ":"
- for every in fileslist:
- outputFileName = os.path.splitext(every)
- print "Generate " + outputFileName[0]
- outputFilePath = os.path.join(base_dir, outputFileName[0])
- generate_script_from_templatefile(every, outputFilePath,target_name, render_dict)
-
-def validate_conv_conv ( layerA, layerB ) :
- if (len(layerA["iDim"]) == 1 or len(layerB["iDim"]) == 1) :
- print "Unexpected: convolution layer with one dimension"
-
- oDim = layerA["iDim"][0]
- if (layerA["pad"] == "same"):
- oDim += (layerA["window"][0] - 1)
- oDim -= layerA["window"][0]
- oDim /= layerA["step"][0]
- oDim += 1
-
- if (oDim != layerB["iDim"][0]):
- print "Incompatible layer: " + layerA["name"] + ", " + layerB["name"]
- print "oDim = " + str(oDim) + ", layerB iDim = " + str(layerB["iDim"][0])
-
- if (layerA["filters"] != layerB["iDim"][2]):
- print "Incompatible layer: " + layerA["name"] + ", " + layerB["name"]
- print "filters = " + str(layerA["filters"]) + " iDim = " + str(layerB["iDim"][2])
-
-def validate_conv_pool ( layerA, layerB ) :
- validate_conv_conv( layerA, layerB) #for now, same implementation as conv_conv
-
-def validate_conv_full ( layerA, layerB ) :
- oDim = layerA["iDim"][0]
- if (layerA["pad"] == "same"):
- oDim += (layerA["window"][0] - 1)
- oDim -= layerA["window"][0]
- oDim /= layerA["step"][0]
- oDim += 1
-
- size = oDim * oDim * layerA["filters"]
-
- if ( size != layerB["iDim"][0]) :
- print "Incompatible layer: " + layerA["name"] + ", " + layerB["name"]
- print "oDim = " + str(oDim) + ", layerB iDim = " + str(layerB["iDim"][0])
-
-def validate_pool_conv ( layerA, layerB) :
- oDim = layerA["iDim"][0]
- if (layerA["pad"] == "same"):
- oDim += (layerA["window"][0] - 1)
- oDim -= layerA["window"][0]
- oDim /= layerA["step"][0]
- oDim += 1
-
- if (oDim != layerB["iDim"][0]):
- print "Incompatible layer: " + layerA["name"] + ", " + layerB["name"]
- print "oDim = " + str(oDim) + ", layerB iDim = " + str(layerB["iDim"][0])
-
- if (layerA["iDim"][2] != layerB["iDim"][2]) :
- print "Incompatible layer: " + layerA["name"] + ", " + layerB["name"]
- print "filters = " + str(layerA["iDim"][2]) + " iDim = " + str(layerB["iDim"][2])
-
-def validate_pool_full( layerA, layerB) :
- oDim = layerA["iDim"][0]
- if (layerA["pad"] == "same"):
- oDim += (layerA["window"][0] - 1)
- oDim -= layerA["window"][0]
- oDim /= layerA["step"][0]
- oDim += 1
-
- size = oDim * oDim * layerA["iDim"][2]
-
- if (size != layerB["iDim"][0]) :
- print "Incompatible layer: " + layerA["name"] + ", " + layerB["name"]
- print "size = " + str(size) + ", layerB iDim = " + str(layerB["iDim"][0])
-
-def validate_full_full( layerA, layerB) :
- if (layerA["filters"] != layerB["iDim"][0]) :
- print "Incompatible layer: " + layerA["name"] + ", " + layerB["name"]
- print "size = " + str(size) + ", layerB iDim = " + str(layerB["iDim"][0])
-
-def validate_layer( layerA, layerB) :
- if (layerA["layer_type"] == "conv"):
- if (layerB["layer_type"] == "conv"):
- validate_conv_conv( layerA, layerB)
- elif (layerB["layer_type"] == "max_pool"):
- validate_conv_pool( layerA, layerB)
- elif (layerB["layer_type"] == "fc"):
- validate_conv_full( layerA, layerB)
- else :
- print "Unsupported Layer connection"
- elif (layerA["layer_type"] == "max_pool"):
- if (layerB["layer_type"] == "conv"):
- validate_pool_conv( layerA, layerB)
- elif (layerB["layer_type"] == "max_pool"):
- print "Unexpected -- two pool layers back to back"
- elif (layerB["layer_type"] == "fc"):
- validate_pool_full( layerA, layerB)
- else:
- print "Unsupported Layer connection"
- elif (layerA["layer_type"] == "fc"):
- if (layerB["layer_type"] == "conv"):
- print "Unsupported layer validation"
- elif (layerB["layer_type"] == "max_pool"):
- print "Unsupported layer validation"
- elif (layerB["layer_type"] == "fc"):
- validate_full_full( layerA, layerB)
- else :
- print "Unsupported Layer connection"
- else :
- print "Unsupported Layer connection"
-
-def main(builtinParameters = {}):
- # parse input arguments
- parser = argparse.ArgumentParser(description='DNNOpt code generator')
- parser.add_argument('spec', help='location of the spec file')
- parser.add_argument('--debug', help='if debugging messages and features are enabled', action='store_true')
- parser.add_argument('--verify', help='specify type of verification, single layer or network', action='store')
- parser.add_argument('--layer_test_method', help='specify method of verification for single layer: random/golden', action='store')
- parser.add_argument('--network_test_method', help='specify method of verification for the network: groundtruth/golden', action='store')
- args = parser.parse_args()
- cwd = os.getcwd()
-
- # read input spec file
- with open(os.path.join(cwd, args.spec), 'r') as jsondata:
- # network_spec = ast.literal_eval(jsondata)
- network_spec = json.load(jsondata)
-
- # parsing arguments
- network_spec['flags'] = dict()
- # if debugging mode is on
- # TODO: improve the error messages
- if(args.debug):
- print("INFO: Debugging mode")
- network_spec['flags']['debug'] = True
- if(args.verify == 'layer'):
- network_spec['flags']['verify'] = 'layer'
- if((args.layer_test_method != 'random') and (args.layer_test_method != 'golden')):
- sys.exit('ERROR: unrecognized test method')
- else:
- network_spec['flags']['test_method'] = args.layer_test_method
- if(len(network_spec['network']) > 1):
- sys.exit('ERROR: more than one layer given for layer verification')
- else:
- print('INFO: Generating verification code for single layer')
- elif(args.verify == 'network'):
- print('INFO: Generating verification code for whole network')
- if((args.network_test_method != 'groundtruth') and (args.network_test_method != 'golden')):
- sys.exit('ERROR: Unrecognized test method')
- else:
- network_spec['flags']['test_method'] = args.network_test_method
- network_spec['flags']['verify'] = 'network'
- elif(args.verify is not None):
- # error handling
- sys.exit("ERROR: Invalid option for validation mode")
- else:
- print("INFO: In clean mode")
-
- # if clustering is enabled
- try:
- cluster_values_file = network_spec['global']['cluster_values_file']
- print("INFO: Clustering enabled, cluster file: " + cluster_values_file)
- with open(os.path.join(cwd, cluster_values_file), 'r') as cluster_file :
- #network_spec = ast.literal_eval(jsondata)
- cluster = json.load(cluster_file)
- except KeyError:
- print("WARNNING: Cluster weight file name not found, clustering disabled")
-
- for idx, layer in enumerate(network_spec['network']):
- layer_name = layer['name']
- if (layer['layer_type'] == 'conv') or (layer['layer_type'] == 'fc'):
- if layer['preprocess']['cluster'] == True:
- network_spec['network'][idx]['cluster_values'] = cluster[layer_name]
-
- #for index in range(len(network_spec["network"])):
- # if (index != 0) :
- # validate_layer(network_spec["network"][index-1], network_spec["network"][index])
-
- scan_for_templatefile(cwd, "dnngen/templates", "syncpp", network_spec)
-
-if __name__=='__main__':
- main()
-
diff --git a/DRED_codegen/templates/aux_funcs.inc b/DRED_codegen/templates/aux_funcs.inc
deleted file mode 100644
index eb081c69216cc7d71b188ac7890294c5f06609ec..0000000000000000000000000000000000000000
--- a/DRED_codegen/templates/aux_funcs.inc
+++ /dev/null
@@ -1,470 +0,0 @@
-{% import 'global_macros.inc' as gm %}
-{% set ident = gm.ident %}
-{%- macro fcall_readW(object) -%}
-read_weights<int8_t>("{{object.weights.W}}.bin", {{object.wshape[0]}}, {{object.wshape[1]}}, {{object.wshape[2]}}, {{object.wshape[3]}}, (ap_int<512>*)w);
-{%- endmacro -%}
-
-{%- macro fcall_readW_fc(object, prev) -%}
-{% if prev.layer_type == 'fc' %}
-read_weights_fc<int8_t>("{{object.weights.W}}.bin", 1, 1, {{prev.wshape[1]}}, {{object.wshape[1]}}, (ap_int<512>*)w);
-{% elif prev.layer_type == 'max_pool' %}
-{%- set stride = prev.weights.stride[1] -%}
-{%- set ksize = prev.weights.ksize[1] -%}
-{%- set padsize = prev.weights.padsize -%}
-{% if prev.mode == 'SAME' %}
-{%- set OH = (prev.dshape[0]/stride)|round(0,'ceil')|int -%}
-{%- set OW = (prev.dshape[1]/stride)|round(0,'ceil')|int -%}
-{% else %}
-{%- set OH = ((prev.dshape[0]-ksize+1)/stride)|round(0,'ceil')|int -%}
-{%- set OW = ((prev.dshape[1]-ksize+1)/stride)|round(0,'ceil')|int -%}
-{% endif %}
-read_weights_fc<int8_t>("{{object.weights.W}}.bin", {{OH}}, {{OW}}, {{prev.dshape[2]}}, {{object.wshape[1]}}, (ap_int<512>*)w);
-{% endif %}
-{%- endmacro -%}
-
-{%- macro fcall_readb(object) -%}
-{%- if object.layer_type == 'conv' %}
-read_bias<int8_t>("{{object.weights.b}}.bin", {{object.wshape[3]}}, {{object.weights.b_width}}, {{object.loop_factor}}, {{object.interface.memory_bit_width}}, (ap_int<512>*)w);
-{%- elif object.layer_type == 'max_pool' -%}
-read_bias<int8_t>("{{object.weights.b}}.bin", {{object.dshape[2]}}, {{object.weights.b_width}}, {{object.loop_factor}}, {{object.interface.memory_bit_width}}, (ap_int<512>*)w);
-{%- elif object.layer_type == 'fc' -%}
-read_bias<int8_t>("{{object.weights.b}}.bin", {{object.dshape[0]}}, {{object.weights.b_width}}, {{object.op_loop_factor}}, {{object.interface.memory_bit_width}}, (ap_int<512>*)w);
-{%- endif -%}
-{%- endmacro -%}
-
-{%- macro fcall_readIp(object) -%}
-read_input("./test_data/test_image.bin", {{object.dshape[0]}}, {{object.dshape[1]}}, {{object.dshape[2]}}, a);
-{%- endmacro -%}
-
-{%- macro fimpl_aux() -%}
-{# ATTENTION: the following parameters might need to be tweaked
- when testing the framework under various bit-width #}
-{% set w_bw = 8 %} {# bit-width of the weights#}
-{% set b_bw = 8 %} {# bit-width of the biases#}
-{% set ip_bw = 8 %}{# bit-width of the input TO THE NETWORK #}
-{% set loop_factor = 32 %} {# temporary setting #}
-{% set mem_bw = 512 %} {# temporary setting #}
-{% set iters_weight = (w_bw*loop_factor*2/mem_bw)|round(0,'ceil')|int %}
-template <class T>
-void display_4d(
- int N,
- int C,
- int H,
- int W,
- T arr[]
-){
-{% call gm.forloop(1, 'co', 'N') %}
-{% call gm.forloop(2, 'ci', 'C') %}
-{% call gm.forloop(3, 'fh', 'H') %}
-{% call gm.forloop(4, 'fw', 'W') %}
-{{gm.ident(5)}}cout << setw(4) << arr[fh*N*C*W+fw*N*C+ci*N+co];
-{%- endcall %}
-{{gm.ident(4)}}cout << endl;
-{%- endcall %}
-{{gm.ident(3)}}cout << endl;
-{%- endcall -%}
-{%- endcall %}
-}
-
-template <class T>
-void display_3d(
- int H,
- int W,
- int C,
- T arr[]
-){
-{% call gm.forloop(1, 'c', 'C') %}
-{% call gm.forloop(2, 'h', 'H') %}
-{% call gm.forloop(3, 'w', 'W') %}
-{{gm.ident(4)}}cout << setw(4) << (int)arr[c*H*W+h*W+w];
-{%- endcall -%}
-{{gm.ident(3)}}cout << endl;
-{%- endcall -%}
-{{gm.ident(2)}}cout << endl;
-{%- endcall -%}
-}
-
-{# Golden model to compare to #}
-template <class T>
-void conv_2d(T output[],
- T kernel[],
- T input[],
- int FH, //Filter height
- int FW, //Filter width
- int IH, //Input height
- int IW, //Input width
- int CI, //Input channels
- int OH, //Output Height(after padding)
- int OW, //Output Width (after padding)
- int CO, //Output Channels
- int S) //Stride
-{
-T temp, local_sum;
-/**************main loop body****************************/
-{% call gm.forloop(1, 'co', 'CO') %}
-{% call gm.forloop(2, 'out_h', 'OH') %}
-{% call gm.forloop(3, 'out_w', 'OW') %}
-{{gm.ident(4)}}local_sum = 0;
-{% call gm.forloop(4, 'ci', 'CI') %}
-{% call gm.forloop(5, 'fh', 'FH') %}
-{% call gm.forloop(6, 'fw', 'FW') %}
-{# Note: The order of dimension for input is CHW and HWCN for the kernel#}
-{{gm.ident(7)}}temp = input[ci*IH*IW+(out_h*S+fh)*IW+(out_w*S+fw)];
-{{gm.ident(7)}}local_sum += kernel[fh*(FW*CI*CO)+fw*(CI*CO)+ci*CO+co]*temp;
-{% endcall %}
-{% endcall %}
-{% endcall %}
-{{gm.ident(4)}}output[co*OH*OW+out_h*OW+out_w] = local_sum;
-{% endcall %}
-{% endcall %}
-{% endcall %}
-}
-
-{# Golden model for Fully Connected Layer, essentially a Mat-Vec Multiplication #}
-template <class T>
-void fc(T output[],
- T weight[],
- T input[],
- int IW,
- int OW)
-{
-{% call gm.forloop(1, 'ow', 'OW') %}
-{% call gm.forloop(2, 'iw', 'IW') %}
-{{gm.ident(3)}}output[ow] += input[iw]*weight[iw*OW+ow];
-{%- endcall -%}
-{%- endcall -%}
-}
-
-{# since there's no need to parameterize these functions
- leave them as is here #}
-template <class T>
-void relu_gold(
- int IH,
- int IW,
- int IC,
- int padding,
- T in[],
- T bias[],
- T out[]
- ){
- int OW = IW + 2 * padding;
- int OH = IH + 2 * padding;
-
- for(int c = 0;c < IC;c ++){
- for(int i = 0;i < padding*OW;i ++){
- out[c*OH*OW+i] = 0;
- }
- for(int h = 0;h < IH;h ++){
- for(int i = 0;i < padding;i ++){
- out[c*OH*OW+(h+padding)*OW+i] = 0;
- }
- for(int w = 0;w < IW;w ++){
- T temp = in[c*IH*IW+h*IW+w];
- if(temp > 0){
- out[c*OH*OW+(h+padding)*OW+w+padding] = temp + bias[c];
- } else {
- out[c*OH*OW+(h+padding)*OW+w+padding] = 0;
- }
- }
- for(int i = 0;i < padding;i ++){
- out[c*OH*OW+(h+padding)*OW+IW+padding+i] = 0;
- }
- }
- for(int i = 0;i < padding*OW;i ++){
- out[c*OH*OW+(IH+padding)*OW+i] = 0;
- }
- }
-}
-
-template <class T>
-void maxpool_gold(
- int IH,
- int IW,
- int IC,
- int padding,
- int k_size,
- int stride,
- bool is_same,
- T in[],
- T bias[],
- T out[]
- ){
- int OW, OH, w_start, w_end, end;
-
- if(is_same){
- OW = ceil((float)IW/(float)stride) + 2 * padding;
- OH = ceil((float)IH/(float)stride) + 2 * padding;
- w_start = -(k_size-1)/2;
- w_end = (k_size-1)/2 + 1;
- end = IW;
- } else {
- OW = ceil((float)(IW - k_size + 1)/(float)stride + 2 * padding);
- OH = ceil((float)(IH - k_size + 1)/(float)stride + 2 * padding);
- w_start = 0;
- w_end = k_size;
- end = IW - k_size + 1;
- }
-
- for(int c = 0;c < IC;c ++){
- for(int i = 0;i < padding*OW;i ++){
- out[c*OH*OW+i] = 0;
- }
- for(int h = 0;h < end;h += stride){
- for(int i = 0;i < padding;i ++){
- out[c*OH*OW+(h/stride+padding)*OW+i] = 0;
- }
- for(int w = 0;w < end;w += stride){
- T rst = 0;
- for(int hh = w_start;hh < w_end;hh ++){
- for(int ww = w_start;ww < w_end;ww ++){
- if (!(((hh+h)<0)||((ww+w)<0)||((hh+h)>=IH)||((ww+w)>=IW))){
- T temp = in[c*IH*IW+(h+hh)*IW+(w+ww)] + bias[c];
- if(temp > rst) rst = temp;
- }
- }
- }
- out[c*OH*OW+(h/stride+padding)*OW+w/stride+padding] = rst;
- }
- for(int i = 0;i < padding;i ++){
- out[c*OH*OW+(h/stride+padding)*OW+OW-padding+i] = 0;
- }
- }
- for(int i = 0;i < padding*OW;i ++){
- out[c*OH*OW+(OH-padding)*OW+i] = 0;
- }
- }
-}
-
-{# Universal data read function, read in data from *.bin #}
-template<class T>
-int read_data(const char *path, T* array, int size){
- FILE *fp;
- fp = fopen(path,"rb");
- if(fp == NULL) {
- printf("Input file not found\n"); abort();
- }
- int ret;
- ret = fread(array,sizeof(T), size, fp);
- fclose(fp);
- return ret;
-}
-
-{# Function that packs the data from a regular array
- This function is also used by the validation functions #}
-template<class T>
-void pack_weights(int FH, int FW, int CII, int COO, T in[], ap_int<{{mem_bw}}>out[50176]){
-{{gm.ident(1)}}ap_int<{{w_bw*loop_factor}}> outPack1 = 0;
-{{gm.ident(1)}}ap_int<{{w_bw*loop_factor}}> outPack2 = 0;
-{% call gm.forloop(1, 'i', '50176') %}
-{{gm.ident(1)}}out[i] = 0;
-{%- endcall %}
-{{gm.ident(1)}}int address = 0;
-{% call gm.forloop(1, 'fh', 'FH') %}
-{% call gm.forloop(2, 'fw', 'FW') %}
-{% call gm.forloop_s(3, 'cii', 'CII', loop_factor) %}
-{{gm.ident(3)}}//input is packed by {{loop_factor}}
-{% call gm.forloop_s(4, 'coo', 'COO', 2) %}
-{% call gm.forloop(5, 'i', loop_factor) %}
-{# Since the computatino requires 2 packs of weight for each loop cycle
- we need to pack 2 packs in each batch
- Notice that we assumes the number of output layers to be even #}
-//Pack data in the array
-{{gm.ident(6)}}if(cii+i < CII){
-{{gm.ident(7)}}outPack1.{{gm.pr_range_idx(w_bw, 'i')}} =
-{{gm.ident(8)}}in[fh*(FW*CII*COO)+fw*(CII*COO)+(cii+i)*COO+coo];
-{{gm.ident(7)}}outPack2.{{gm.pr_range_idx(w_bw, 'i')}} =
-{{gm.ident(8)}}in[fh*(FW*CII*COO)+fw*(CII*COO)+(cii+i)*COO+coo+1];
-{{gm.ident(6)}}} else {
-{{gm.ident(7)}}outPack1.{{gm.pr_range_idx(w_bw, 'i')}} = 0; //pad 0
-{{gm.ident(7)}}outPack2.{{gm.pr_range_idx(w_bw, 'i')}} = 0; //pad 0
-{{gm.ident(6)}}}
-{%- endcall -%}
-{{gm.ident(5)}}//write pack to the buffer according to mem bitwidth
-{%for idx in range(0,iters_weight)%}
-{# if this is not the last pack#}
-{% if ((idx+1)*(mem_bw/2) < w_bw*loop_factor) %}
-{{gm.ident(5)}}out[address+{{idx}}].{{gm.pr_range_bound((mem_bw/2-1)|int,0)}} = outPack1.{{gm.pr_range_bound(((idx+1)*(mem_bw/2)-1)|int,(idx*mem_bw/2)|int)}};
-{{gm.ident(5)}}out[address+{{idx}}].{{gm.pr_range_bound(mem_bw-1,(mem_bw/2)|int)}} = outPack2.{{gm.pr_range_bound(((idx+1)*(mem_bw/2)-1)|int,(idx*mem_bw/2)|int)}};
-{# if this is the last pack#}
-{% else %}
-{{gm.ident(5)}}out[address+{{idx}}].{{gm.pr_range_bound(w_bw*loop_factor-1-(idx*mem_bw/2)|int,0)}} = outPack1.{{gm.pr_range_bound(w_bw*loop_factor-1,(idx*mem_bw/2)|int)}};
-{{gm.ident(5)}}out[address+{{idx}}].{{gm.pr_range_bound(w_bw*loop_factor-1-((idx-1)*mem_bw/2)|int,(mem_bw/2)|int)}} = outPack2.{{gm.pr_range_bound(w_bw*loop_factor-1,(idx*mem_bw/2)|int)}};
-{%endif%}
-{%endfor%}
-{{gm.ident(5)}}address += {{iters_weight}};
-{%- endcall -%}
-{%- endcall -%}
-{%- endcall -%}
-{%- endcall -%}
-}
-
-template<class T>
-void read_weights(const char* path, int FH, int FW, int CII, int COO, ap_int<{{mem_bw}}>out[50176]){
-{{gm.ident(1)}}T in[FH*FW*CII*COO];
-{{gm.ident(1)}}read_data<T>(path, (T*)in, FH*FW*CII*COO);
-{{gm.ident(1)}}pack_weights<T>(FH, FW, CII, COO, in, out);
-}
-
-{# this function is parameterized with runtime parameters,
- so it does not need to be parameterized in template #}
-template<class T>
-void pack_input(int H, int W, int C, int padSize, int packSize, int bitWidth,T in[], ap_int<512> out[50176]){
- int OH = H + 2*padSize;
- int OW = W + 2*padSize;
- int numBufLines = ceil((float)(packSize*bitWidth)/(float)512); //512 is the bit-width of one buffer line
- int numbersPerLine = 512/bitWidth;
-
- for (int c=0; c < C; c++) {
- int packIdx = c / packSize;
- int lineIdx = (c % packSize)/numbersPerLine;
- int linePos = (c % packSize)%numbersPerLine;
- //pad leading zero rows
- for (int i=0; i < padSize*OW*numBufLines; i++) {
- out[(c*OH*OW)*numBufLines+i]=0;
- }
- for (int h=0; h < H; h++) {
- // pad zeros in front of each line
- for (int i=0; i < padSize*numBufLines; i++) {
- out[(c*OH*OW+(padSize+h)*OW)*numBufLines+i]=0;
- }
- for (int w=0; w < W; w++) {
- out[(packIdx*OH*OW+(padSize+h)*OW+padSize+w)*numBufLines+lineIdx].range(bitWidth*linePos+bitWidth-1,bitWidth*linePos) =
- in[h*W*C+w*C+c];
- }
- //pad zeros at the end of each line
- for (int i=0; i < padSize*numBufLines; i++) {
- out[(c*OH*OW+(padSize+h)*OW+W+padSize)*numBufLines+i]=0;
- }
-
- }
- for (int i=0; i < padSize*OW*numBufLines; i++) {
- out[(c*OH*OW+(padSize+H)*OW)*numBufLines+i]=0;
- }
-
- }
-}
-
-template<class T>
-void read_input(const char* path, int IH, int IW, int CII, int pad, ap_int<512> out[50176]){
-{{gm.ident(1)}}T in[CII*IH*IW];
-{{gm.ident(1)}}read_data<T>(path, (T*)in, IH*IW*CII);
-{{gm.ident(1)}}//display_3d<T>(IH, IW, CII, in);
-{{gm.ident(1)}}/* ATTENTION:
-{{gm.ident(1)}} * pad size set to 2 for LeNet, remember to change this for other nets
-{{gm.ident(1)}} * input loop factor is set to 32, bitWidth of input is 8
-{{gm.ident(1)}} */
-{{gm.ident(1)}}pack_input<T>(IH, IW, CII, pad, 32, 8, in, out);
-}
-
-template<class T>
-void pack_weights_fc(int IH, int IW, int CII, int COO, T in[], ap_int<512>out[50176]){
-{{gm.ident(1)}}int address = 0;
-{{gm.ident(1)}}ap_int<{{w_bw*loop_factor}}> outPack1 = 0;
-{{gm.ident(1)}}ap_int<{{w_bw*loop_factor}}> outPack2 = 0;
-{% call gm.forloop(1, 'i', '50176') %}
-{{gm.ident(1)}}out[i] = 0;
-{%- endcall %}
-{% call gm.forloop_s(1, 'cii', 'CII', loop_factor) %}
-{% call gm.forloop(2, 'h', 'IH') %}
-{% call gm.forloop(3, 'w', 'IW') %}
-{{gm.ident(3)}}//input is packed by {{loop_factor}}
-{% call gm.forloop_s(4, 'coo', 'COO', 2) %}
-{% call gm.forloop(5, 'ci', loop_factor) %}
-{# Same as read_weights function for conv above#}
-{{gm.ident(6)}}if(cii+ci < CII){
-{{gm.ident(7)}}outPack1.{{gm.pr_range_idx(w_bw, 'ci')}}=in[h*CII*COO*IW+w*CII*COO+(cii+ci)*COO+coo];
-{{gm.ident(7)}}outPack2.{{gm.pr_range_idx(w_bw, 'ci')}}=in[h*CII*COO*IW+w*CII*COO+(cii+ci)*COO+coo+1];
-{{gm.ident(6)}}} else {
-{{gm.ident(7)}}outPack1.{{gm.pr_range_idx(w_bw, 'ci')}} = 0;
-{{gm.ident(7)}}outPack2.{{gm.pr_range_idx(w_bw, 'ci')}} = 0;
-{{gm.ident(6)}}}
-{%- endcall -%}
-{{gm.ident(5)}}//write pack
-{% set iters_weight = (w_bw*loop_factor*2/mem_bw)|round(0,'ceil')|int %}
-{%for idx in range(0,iters_weight)%}
-{# if this is not the last pack#}
-{% if ((idx+1)*(mem_bw/2) < w_bw*loop_factor) %}
-{{gm.ident(5)}}out[address+{{idx}}].{{gm.pr_range_bound((mem_bw/2-1)|int,0)}} = outPack1.{{gm.pr_range_bound(((idx+1)*(mem_bw/2)-1)|int,(idx*mem_bw/2)|int)}};
-{{gm.ident(5)}}out[address+{{idx}}].{{gm.pr_range_bound(mem_bw-1,(mem_bw/2)|int)}} = outPack2.{{gm.pr_range_bound(((idx+1)*(mem_bw/2)-1)|int,(idx*mem_bw/2)|int)}};
-{# if this is the last pack#}
-{% else %}
-{{gm.ident(5)}}out[address+{{idx}}].{{gm.pr_range_bound(w_bw*loop_factor-1-(idx*mem_bw/2)|int,0)}} = outPack1.{{gm.pr_range_bound(w_bw*loop_factor-1,(idx*mem_bw/2)|int)}};
-{{gm.ident(5)}}out[address+{{idx}}].{{gm.pr_range_bound(w_bw*loop_factor-1-((idx-1)*mem_bw/2)|int,(mem_bw/2)|int)}} = outPack2.{{gm.pr_range_bound(w_bw*loop_factor-1,(idx*mem_bw/2)|int)}};
-{%endif%}
-{%endfor%}
-{{gm.ident(5)}}address += {{iters_weight}};
-{% endcall -%}
-{% endcall -%}
-{% endcall -%}
-{% endcall -%}
-}
-
-template <class T>
-void read_weights_fc(const char* path, int IH, int IW, int CII, int COO, ap_int<512>out[50176]){
-{{gm.ident(1)}}T in[IH*IW*CII*COO]; //notice that IW IH CII are for the layer before flatten for convenience
-{{gm.ident(1)}}read_data<T>(path, (T*)in, IH*IW*CII*COO);
-{{gm.ident(1)}}pack_weights_fc(IH, IW, CII, COO, in, out);
-}
-
-template <class T>
-void read_bias(const char* path, int COO, int bw, int loop_factor, int mem_bw, ap_int<512>out[50176]){
- T array[COO];
- ap_int<1024> outPack = 0; //a relatively big buffer
- read_data<T>(path, (T*)array, COO);
-
- for (int i=0;i<50176;i++){
- out[i] = 0;
- }
- int address = 0;
- for(int coo=0;coo<COO;coo+=loop_factor){
- for(int i=0;i<loop_factor;i++){
- if(coo+i < COO){
- outPack.range(bw-1+i*bw,i*bw) = array[coo+i];
- } else {
- outPack.range(bw-1+i*bw,i*bw) = 0; //pad 0
- }
- }
- int iters_bias = ceil((float)(bw*loop_factor)/(float)mem_bw);
- for(int i = 0;i < iters_bias;i ++){
- if((i+1)*mem_bw <= bw*loop_factor){
- out[address].range(mem_bw-1,0) = outPack.range((i+1)*mem_bw-1,i*mem_bw);
- } else {
- out[address].range(bw*loop_factor-i*mem_bw-1,0) = outPack.range(bw*loop_factor-1, i*mem_bw);
- }
- address ++;
- }
- }
-}
-
-template<int mem_bw, int bw>
-bool compare_almost_equal(int H, int W, int C, int loop_factor, int ref[], xcelo_uint<512> real[]){
- int iters = ceil((float)(bw*loop_factor)/(float)mem_bw);
- int num_per_pack = mem_bw/bw;
- //CC is number of channels after packing
- int CC = ceil((float)C/(float)loop_factor);
- bool flag = false;
-
- for(int cc = 0;cc < CC;cc ++){
- for(int i = 0;i < loop_factor;i ++){
- for(int h = 0;h < H;h ++){
- for(int w = 0;w < W;w ++){
- //NOTE: assume mem_bw is 512 for now, might change this value in the future
- ap_int<mem_bw> temp = ((ap_int<mem_bw>*)real)[(cc*H*W+h*W+w)*iters];
- if((cc*loop_factor + i) < C){
- ap_int<mem_bw> temp = ((ap_int<mem_bw>*)real)[(cc*H*W+h*W+w)*iters + i/num_per_pack];
- ap_int<bw> temp_real = temp.range(i*bw+(bw-1), i*bw);
- ap_int<bw> temp_ref = ref[(cc*loop_factor+i)*H*W+h*W+w];
- //NOTE: for now fix the error threshold at 2, might decide to make this configurable in the future
- if(abs(temp_real - temp_ref) > 2){
- cout << "[" << (cc*loop_factor+i) << "][" << h << "][" << w << "]";
- cout << "Diff: Real--" << temp_real << " Ref--" << temp_ref << endl;
- flag = true;
- }
- }
- }
- }
- }
- }
- return flag;
-}
-{% endmacro %}
\ No newline at end of file
diff --git a/DRED_codegen/templates/avg_pool.inc b/DRED_codegen/templates/avg_pool.inc
deleted file mode 100644
index f2a060ecb1f664c1e989140831e913fc1ea3fe03..0000000000000000000000000000000000000000
--- a/DRED_codegen/templates/avg_pool.inc
+++ /dev/null
@@ -1,35 +0,0 @@
-{%- macro handle_preprocess(object, name) -%}
- {%- if name in object.keys() %}
- {%- for key, value in object.iteritems() %}
- {%- if key == name %}
- {%- if value > 0 %}
- {%- endif %}
- {%- endif %}
- {%- endfor %}
- {%- endif %}
-{%- endmacro %}
-
-{%- macro fdecl(object) -%}
-void {{object.name}}_avg_pool(arguments);
-{%- endmacro %}
-
-{%- macro var_decl(object) -%}
- {%- if object.preprocess %}
- {%- else %}
- {%- endif %}
-{%- endmacro %}
-
-{%- macro fvalid_call(object) %}
-{%- endmacro %}
-
-{%- macro fcall(object) -%}
-{{object.name}}_avg_pool();
-{%- endmacro %}
-
-{%- macro fimpl(object) -%}
-void {{object.name}}_avg_pool(arguments) {};
-{%- endmacro %}
-
-
-{%- macro fvalid(object) -%}
-{%- endmacro %}
\ No newline at end of file
diff --git a/DRED_codegen/templates/batch_norm.inc b/DRED_codegen/templates/batch_norm.inc
deleted file mode 100644
index abc4bc0ac6d31e0ee7874f00d3235ffaaa55e12a..0000000000000000000000000000000000000000
--- a/DRED_codegen/templates/batch_norm.inc
+++ /dev/null
@@ -1,161 +0,0 @@
-{% import "global_macros.inc" as gm %}
-{% import "aux_funcs.inc" as aux %}
-
-{%- macro handle_preprocess(object, name) -%}
- {%- if name in object.keys() %}
- {%- for key, value in object.iteritems() %}
- {%- if key == name %}
- {%- if value > 0 %}
- {%- endif %}
- {%- endif %}
- {%- endfor %}
- {%- endif %}
-{%- endmacro %}
-
-{%- macro fdecl(object) -%}
-{%- endmacro %}
-
-{%- macro var_decl(object) -%}
- {%- if object.preprocess %}
- {%- else %}
- {%- endif %}
-{%- endmacro %}
-
-{% macro fvalid_call(object) %}
-{{gm.ident(1)}}{{object.name}}_test_pool();
-{{gm.ident(1)}}{{object.name}}_test_pool_flatten();
-{% endmacro %}
-
-{# defines paramters passed in when it get called#}
-{%- macro fcall(object, debug) -%}
-{%- set loop_factor = object.loop_factor -%}
-{%- set ip_shift = object.preprocess.ip_shift-%}
-{%- set channels = (object.dshape[2]/loop_factor)|round(0,'ceil')|int-%}
-{{gm.ident(1)}}{{object.name}}_layer ({{object.dshape[0]}}, {{object.dshape[1]}}, {{channels}}, {{object.weights.mean}}, {{object.weights.dev_inv}}, {{object.weights.gamma}}, {{object.weights.beta}},
-{%- if object.interface.in == 'dram' %}
- a,
-{%- elif object.interface.in == 'onchip' %}
- a_onchip,
-{%- endif %}
-{%- if object.interface.int == 'dram' %}
- c
-{%- elif object.interface.int == 'onchip' %}
- c_onchip
-{%- endif %}
-);
-{%- endmacro %}
-
-{% macro readFunc(layer, name, mem, bw, loop_factor)%}
-{%- set iters = (bw*loop_factor/mem)|round(0,'ceil')|int -%}
-int read{{name}}_{{layer}}({{gm.fixed_int(mem)}} *mem,
- {{gm.fixed_int(bw*loop_factor)}} &input,
- int address)
-{
- int index=0;
- //Read requires {{iters}} iterations
-{%for idx in range(0,iters)%}
-{# if this is not the last pack#}
-{%if ((idx+1)*mem < bw*loop_factor) %}
- input.{{gm.pr_range_bound((idx+1)*mem-1,idx*mem)}}=mem[address+index].{{gm.pr_range_bound((mem-1),0)}};
- index ++;
-{# if this is the last pack#}
-{% else %}
- input.{{gm.pr_range_bound(bw*loop_factor-1,idx*mem)}}=mem[address+index].{{gm.pr_range_bound(bw*loop_factor-idx*mem-1,0)}};
-{%endif%}
-{%endfor%}
- return address;
-}
-{% endmacro %}
-
-{% macro writeFunc(layer, name, mem, bw, loop_factor)%}
-{%- set iters = (bw*loop_factor/mem)|round(0,'ceil')|int -%}
-int write{{name}}_{{layer}}({{gm.fixed_int(mem)}} *mem,
- {{gm.fixed_int(bw*loop_factor)}} output,
- int address)
-{
- int index=0;
- //Write requires {{iters}} iterations
-{%for idx in range(0,iters)%}
-{# if this is not the last pack#}
-{%if ((idx+1)*mem < bw*loop_factor) %}
- mem[address+index].{{gm.pr_range_bound((mem-1),0)}} = output.{{gm.pr_range_bound((idx+1)*mem-1,idx*mem)}};
- index ++;
-{# if this is the last pack#}
-{% else %}
- mem[address+index].{{gm.pr_range_bound(bw*loop_factor-idx*mem-1,0)}}=output.{{gm.pr_range_bound(bw*loop_factor-1,idx*mem)}};
-{%endif%}
-{%endfor%}
- return address;
-}
-{% endmacro %}
-
-{%- macro fimpl(object) -%}
-{%- set ip_bw = object.preprocess.ip_bit_width-%}
-{%- set ip_shift = object.preprocess.ip_shift-%}
-{%- set op_bw = object.preprocess.op_bit_width-%}
-{%- set op_shift = object.preprocess.op_shift -%}
-{%- set loop_factor = object.loop_factor -%}
-{%- set mem_bw = object.interface.memory_bit_width -%}
-{# internal bw is the "safe" bit width for the result of multiplication #}
-{%- set iters_input = (ip_bw*loop_factor/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_output = (op_bw*loop_factor/mem_bw)|round(0,'ceil')|int -%}
-
-{{readFunc(object.name, "Input", mem_bw, ip_bw, loop_factor)}}
-{{writeFunc(object.name, "Output", mem_bw, op_bw, loop_factor)}}
-
-void {{object.name}}_kernel (
-{{gm.ident(1)}}{{gm.f_vdecl_arr(ip_bw, ip_bw, 'input', loop_factor)}},
-{{gm.ident(1)}}{{gm.f_vdecl_arr(ip_bw, ip_bw, 'output', loop_factor)}},
-{{gm.ident(1)}}{{gm.f_vdecl(ip_bw, ip_bw, 'mean')}},
-{{gm.ident(1)}}{{gm.f_vdecl(ip_bw, ip_bw, 'dev_inv')}},
-{{gm.ident(1)}}{{gm.f_vdecl(ip_bw, ip_bw, 'gamma')}},
-{{gm.ident(1)}}{{gm.f_vdecl(ip_bw, ip_bw, 'beta')}})
-{
-{{gm.ident(1)}}{{gm.f_vdecl_arr(ip_bw, ip_bw, 'numerator', loop_factor)}};
-{{gm.ident(1)}}{{gm.f_vdecl_arr(ip_bw, ip_bw, 'quotient', loop_factor)}};
-
-{{gm.ident(1)}}// Here dev_inv is 1/sqrt(std.deviation + epsilon)
-{{gm.ident(1)}}// this value can be taken as std.deviation, we need to calculate this outside this module and then pass it to here
-{{gm.ident(1)}}// because it involves sqrt operation and division.
-{{gm.ident(1)}}// the multiplication with dev_inv is equivalent to division by sqrt(std.deviation + epsilon)
-{% call gm.forloop(1, 'i', loop_factor) %}
-{{gm.ident(2)}}numerator[i] = input[i] - mean;
-{{gm.ident(2)}}quotient[i] = (numerator[i]*dev_inv); //DATA RANGE ERROR!!
-{{gm.ident(2)}}output[i] = gamma*(quotient[i])+beta;
-{% endcall %}
-}
-
-void {{object.name}}_layer(
-{{gm.ident(1)}}int IH,
-{{gm.ident(1)}}int IW,
-{{gm.ident(1)}}int IC,
-{{gm.ident(1)}}{{gm.f_vdecl(ip_bw, ip_bw, 'mean')}},
-{{gm.ident(1)}}{{gm.f_vdecl(ip_bw, ip_bw, 'dev_inv')}},
-{{gm.ident(1)}}{{gm.f_vdecl(ip_bw, ip_bw, 'gamma')}},
-{{gm.ident(1)}}{{gm.f_vdecl(ip_bw, ip_bw, 'beta')}},
-{{gm.ident(1)}}ap_int<{{mem_bw}}>input[50176],
-{{gm.ident(1)}}ap_int<{{mem_bw}}>output[50176])
-{
-{{gm.ident(1)}}{{gm.f_vdecl_arr(ip_bw, ip_bw, 'inTemp', loop_factor)}};
-{{gm.ident(1)}}{{gm.f_vdecl_arr(ip_bw, ip_bw, 'outTemp', loop_factor)}};
-{{gm.ident(1)}}{{gm.fi_vdecl(ip_bw*loop_factor, 'inPackTemp')}};
-{{gm.ident(1)}}{{gm.fi_vdecl(mem_bw, 'outPackTemp')}};
-{% call gm.forloop(1, 'i', 'IH*IW*IC') %} //FLATTENED
-{{gm.ident(2)}}readInput_{{object.name}}(input,inPackTemp,i*{{iters_input}});
-{% call gm.forloop(2, 'j', loop_factor) %} {# assume loop factor is the same for all BN layers#}
-{{gm.ident(3)}}inTemp[i].{{gm.pr_range(ip_bw)}}=inPackTemp.{{gm.pr_range_idx(ip_bw, 'j')}};
-{%- endcall -%}
-{{gm.ident(2)}}{{object.name}}_kernel(inTemp, outTemp, mean, dev_inv, gamma, beta);// perform batch normalization
-{% call gm.forloop(2, 'j', loop_factor) %}
-{{gm.ident(3)}}outPackTemp.{{gm.pr_range_idx(ip_bw, 'j')}} = outTemp[i].{{gm.pr_range(ip_bw)}};
-{%- endcall -%}
-{{gm.ident(2)}}writeOutput_{{object.name}}(output,outPackTemp,i*{{iters_output}});
-{%- endcall -%}
-}
-{%- endmacro -%}
-
-{%- macro fvalid(object) -%}
-void {{object.name}}_test_BN() {
-//Not yet implemented
-}
-{%- endmacro %}
\ No newline at end of file
diff --git a/DRED_codegen/templates/conv.jinja b/DRED_codegen/templates/conv.jinja
deleted file mode 100644
index 2accb3f8a23d4e61bbfdb9f718c90f6edb5d3e84..0000000000000000000000000000000000000000
--- a/DRED_codegen/templates/conv.jinja
+++ /dev/null
@@ -1,278 +0,0 @@
-{% import 'global_macros.inc' as gm %}
-{% import 'aux_funcs.inc' as aux %}
-
-{%- macro gen_red_tree_vdecl(size, dtype) %}
-{%- for level in range(size, 1, -1) %}
-{% if (size % level) == 0 %}
-{{gm.ident(1)}}{{dtype}} inter{{level}}[2][{{level}}];
-{{gm.ident(1)}}#pragma HLS ARRAY_PARTITION variable=inter{{level}} complete dim=1
-{{gm.ident(1)}}#pragma HLS ARRAY_PARTITION variable=inter{{level}} complete dim=2
-{% endif %}
-{%- endfor %}
-{%- endmacro %}
-
-{%- macro gen_red_tree_impl(size, dtype) %}
-{%- for level in range(size, 1, -1) %}
-{%- if (size % level) == 0 %}
-{% if (level != size) %}
-{{gm.ident(1)}}for (int i=0; i<{{level}}; i++) {
-{{gm.ident(2)}}#pragma HLS unroll
-{{gm.ident(2)}}inter{{level}}[0][i]=inter{{level*2}}[0][i*2]+inter{{level*2}}[0][i*2+1];
-{{gm.ident(2)}}inter{{level}}[1][i]=inter{{level*2}}[1][i*2]+inter{{level*2}}[1][i*2+1];
-{{gm.ident(1)}}}
-{%- endif %}
-{%- endif %}
-{%- endfor %}
-{%- endmacro %}
-
-{% macro readFunc(layer, name, mem, bw, loop_factor)%}
-{# Number of packs required to read one pack#}
-{%- set iters = (bw*loop_factor/mem)|round(0,'ceil')|int -%}
-int read{{name}}_{{layer}}({{gm.fixed_int(mem)}} *mem,
- {{gm.fixed_int(bw*loop_factor)}} &input,
- int address)
-{
- int index=0;
- //Read requires {{iters}} iterations
-{%for idx in range(0,iters)%}
-{# if this IS NOT the last pack#}
-{%if ((idx+1)*mem < bw*loop_factor) %}
- input.{{gm.pr_range_bound((idx+1)*mem-1,idx*mem)}}=mem[address+index].{{gm.pr_range_bound((mem-1),0)}};
- index ++;
-{# if this IS the last pack#}
-{% else %}
- input.{{gm.pr_range_bound(bw*loop_factor-1,idx*mem)}}=mem[address+index].{{gm.pr_range_bound(bw*loop_factor-idx*mem-1,0)}};
-{%endif%}
-{%endfor%}
- return address;
-}
-{% endmacro %}
-
-{# Macro that implement the memory write & pack function #}
-{% macro writeFunc(layer, name, mem, bw, loop_factor)%}
-{%- set iters = (bw*loop_factor/mem)|round(0,'ceil')|int -%}
-int write{{name}}_{{layer}}({{gm.fixed_int(mem)}} *mem,
- {{gm.fixed_int(bw*loop_factor)}} output,
- int address)
-{
- int index=0;
- //Write requires {{iters}} iterations
-{%for idx in range(0,iters)%}
-{# if this IS NOT the last pack#}
-{%if ((idx+1)*mem < bw*loop_factor) %}
- mem[address+index].{{gm.pr_range_bound((mem-1),0)}} = output.{{gm.pr_range_bound((idx+1)*mem-1,idx*mem)}};
- index ++;
-{# if this IS the last pack#}
-{% else %}
- mem[address+index].{{gm.pr_range_bound(bw*loop_factor-idx*mem-1,0)}}=output.{{gm.pr_range_bound(bw*loop_factor-1,idx*mem)}};
-{%endif%}
-{%endfor%}
- return address;
-}
-{% endmacro %}
-
-{%- set ip_bw = object.preprocess.ip_bit_width-%}
-{%- set ip_shift = object.preprocess.ip_shift-%}
-{%- set int_bw = object.preprocess.int_bit_width-%}
-{%- set int_shift = object.preprocess.int_shift-%}
-{%- set w_bw = object.weights.W_width-%}
-{%- set w_shift = object.weights.W_shift-%}
-
-{%- set ip_loop_factor = object.ip_loop_factor -%}
-{%- set op_loop_factor = object.op_loop_factor -%}
-{%- set op_unroll_factor = object.op_unroll_factor -%}
-{%- set converge_unroll_factor = object.converge_unroll_factor-%}
-{%- set compengine_unroll_factor = object.compengine_unroll_factor -%}
-
-
-{%- set IH = object.IH -%}
-{%- set IW= object.IW -%}
-{%- set CII = object.CII -%}
-{%- set OH = object.OH -%}
-{%- set OW = object.OW -%}
-{%- set COO = object.COO -%}
-{%- set FH = object.FH -%}
-{%- set FW = object.FW -%}
-{%- set stride = object.stride -%}
-
-{%- set mem_bw = object.memory_bit_width -%}
-{%- set internal_bw = w_bw+ip_bw -%}
-
-{%- set iters_weight = (w_bw*ip_loop_factor*2/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_bias = (ip_bw*ip_loop_factor/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_input = (ip_bw*ip_loop_factor/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_itrm = (int_bw*op_loop_factor/mem_bw)|round(0,'ceil')|int -%}
-
-{# Implement the memory read & unpack function #}
-{# see the overview doc for more detail #}
-
-#include "ap_int.h"
-
-int {{object.name}}_read_weight({{gm.fixed_int(mem_bw)}} *mem,
- {{gm.fixed_int(w_bw*ip_loop_factor)}} &weight1,
- {{gm.fixed_int(w_bw*ip_loop_factor)}} &weight2,
- int address)
-{
- int index=0;
- //Input loop factor is {{ip_loop_factor}}
- //Length of pack is {{w_bw*ip_loop_factor}}
- //Read requires {{iters_weight}} iterations
-{%for idx in range(0,iters_weight)%}
-{# if this is not the last pack#}
-{%if ((idx+1)*(mem_bw/2) < w_bw*ip_loop_factor) %}
- weight1.{{gm.pr_range_bound(((idx+1)*(mem_bw/2)-1)|int,(idx*mem_bw/2)|int)}}=mem[address+index].{{gm.pr_range_bound((mem_bw/2-1)|int,0)}};
- weight2.{{gm.pr_range_bound(((idx+1)*(mem_bw/2)-1)|int,(idx*mem_bw/2)|int)}}=mem[address+index].{{gm.pr_range_bound(mem_bw-1,(mem_bw/2)|int)}};
- index+=1;
-{# if this is the last pack#}
-{% else %}
- weight1.{{gm.pr_range_bound(w_bw*ip_loop_factor-1,(idx*mem_bw/2)|int)}}=mem[address+index].{{gm.pr_range_bound(w_bw*ip_loop_factor-1-(idx*mem_bw/2)|int,0)}};
- weight2.{{gm.pr_range_bound(w_bw*ip_loop_factor-1,(idx*mem_bw/2)|int)}}=mem[address+index].{{gm.pr_range_bound(w_bw*ip_loop_factor-1-((idx-1)*mem_bw/2)|int,(mem_bw/2)|int)}};
-{%endif%}
-{%endfor%}
- return address;
-}
-
-{# Declare Read functions for this layer#}
-{{readFunc(object.name, "Input", mem_bw, ip_bw, ip_loop_factor)}}
-{{readFunc(object.name,"Itrm", mem_bw, int_bw, op_loop_factor)}}
-{{writeFunc(object.name,"Itrm", mem_bw, int_bw, op_loop_factor)}}
-
-void {{object.name}}_pack_dsp(
- {{gm.f_vdecl(internal_bw, internal_bw,'&rst1')}},
- {{gm.f_vdecl(internal_bw, internal_bw,'&rst2')}},
- {{gm.fi_vdecl(w_bw, 'a')}},
- {{gm.fi_vdecl(w_bw, 'b')}},
- {{gm.fi_vdecl(ip_bw, 'c')}}
-) {
- {{gm.fi_vdecl(18+w_bw, 'ai')}} = 0;
- {{gm.fi_vdecl(w_bw, 'bi')}};
- {{gm.fi_vdecl(ip_bw, 'ci')}};
- {{gm.fi_vdecl(45, 'rst')}};
- {{gm.fui_vdecl(19, 'cst')}};
-
- #pragma HLS RESOURCE variable=rst core=DSP48
- #pragma HLS pipeline II=1
- ai.{{gm.pr_range(w_bw)}}=a.{{gm.pr_range(w_bw)}};
- bi.{{gm.pr_range(w_bw)}}=b.{{gm.pr_range(w_bw)}};
- ci.{{gm.pr_range(ip_bw)}}=c.{{gm.pr_range(ip_bw)}};
-
- if ( (ci!=0) && (bi!=0) && (ci.sign() != bi.sign())) {
- cst=0x40000;
- } else {
- cst=0;
- }
- rst = ( (ai<<18) + bi) * ci + cst;
- rst1.{{gm.pr_range(internal_bw)}}=rst.{{gm.pr_range_bound(18+internal_bw-1,18)}};
- rst2.{{gm.pr_range(internal_bw)}}=rst.{{gm.pr_range(internal_bw)}};
-}
-
-void {{object.name}}_out_converge({{gm.fi_vdecl(op_loop_factor*int_bw, '&outpack')}}, {{gm.f_vdecl_arr(internal_bw,internal_bw,'out_temp',op_loop_factor)}}, int right_shift) {
- #pragma HLS pipeline
- #pragma HLS ARRAY_PARTITION variable=out_temp complete dim=1
- {{gm.f_vdecl_arr(int_bw, int_bw,'temp_1',op_loop_factor)}};
- {{gm.f_vdecl_arr(internal_bw, internal_bw,'temp_2',op_loop_factor)}};
- #pragma HLS ARRAY_PARTITION variable=temp_1 complete dim=1
- #pragma HLS ARRAY_PARTITION variable=temp_2 complete dim=1
-{% call gm.forloop(1,'i',op_loop_factor) %}
-{{gm.ident(2)}}#pragma HLS unroll factor={{converge_unroll_factor}}
-{{gm.ident(2)}}temp_1[i].{{gm.pr_range(int_bw)}}=outpack.{{gm.pr_range_idx(int_bw, 'i')}};
-{{gm.ident(2)}}temp_2[i].{{gm.pr_range(internal_bw)}}=out_temp[i].{{gm.pr_range(internal_bw)}};
-{{gm.ident(2)}}temp_1[i]=temp_1[i] + (temp_2[i] >> right_shift);
-{%- endcall -%}
-{% call gm.forloop(1,'i',op_loop_factor) %}
-{{gm.ident(2)}}#pragma HLS unroll factor={{converge_unroll_factor}}
-{{gm.ident(2)}}outpack.{{gm.pr_range_idx(int_bw, 'i')}}=temp_1[i].{{gm.pr_range(int_bw)}};
-{%- endcall -%}
-}
-
-void {{object.name}}_compengine_row(
- {{gm.f_vdecl(internal_bw,internal_bw,'&out1')}},
- {{gm.f_vdecl(internal_bw,internal_bw,'&out2')}},
- {{gm.fi_vdecl(ip_loop_factor*ip_bw, 'InPack')}},
- {{gm.fi_vdecl(w_bw*ip_loop_factor, 'weightPack1')}},
- {{gm.fi_vdecl(w_bw*ip_loop_factor, 'weightPack2')}}
-) {
- {{gm.fi_vdecl_arr(w_bw, 'weight1', ip_loop_factor)}};
- #pragma HLS ARRAY_PARTITION variable=weight1 complete dim=1
- {{gm.fi_vdecl_arr(w_bw, 'weight2', ip_loop_factor)}};
- #pragma HLS ARRAY_PARTITION variable=weight2 complete dim=1
- {{gm.fi_vdecl_arr(ip_bw, 'in', ip_loop_factor)}};
- #pragma HLS ARRAY_PARTITION variable=in complete dim=1
- {{gen_red_tree_vdecl(ip_loop_factor, gm.fixed(internal_bw,internal_bw))}}
- #pragma HLS pipeline II=2
-{% call gm.forloop(1, 'i', ip_loop_factor) %}
-{{gm.ident(2)}}in[i].{{gm.pr_range(ip_bw)}}=InPack.{{gm.pr_range_idx(ip_bw,'i')}};
-{{gm.ident(2)}}weight1[i].{{gm.pr_range(w_bw)}}=weightPack1.{{gm.pr_range_idx(w_bw,'i')}};
-{{gm.ident(2)}}weight2[i].{{gm.pr_range(w_bw)}}=weightPack2.{{gm.pr_range_idx(w_bw,'i')}};
-{%- endcall %}
-{%- call gm.forloop(1, 'i', ip_loop_factor) %}
-{{gm.ident(2)}}#pragma HLS unroll factor={{compengine_unroll_factor}}
-{{gm.ident(2)}}{{object.name}}_pack_dsp(inter{{ip_loop_factor}}[0][i],inter{{ip_loop_factor}}[1][i],weight1[i],weight2[i],in[i]);
-{%- endcall %}
-{{gen_red_tree_impl(ip_loop_factor, gm.fixed(internal_bw,internal_bw))}}
- out1=inter2[0][0]+inter2[0][1];
- out2=inter2[1][0]+inter2[1][1];
-}
-
-
-void {{object.name}}_layer(
- /*
- int IW, //image width with padding
- int IH, //image height with padding
- int CII,//image channels
- int OW, //output width
- int OH, //output height
- int COO, //output channels
- int FW, //filter width
- int FH, //filter height
- int stride,
- */
- int right_shift,
- {{gm.fi_vdecl(mem_bw, 'weight[100352]')}},
- {{gm.fi_vdecl(mem_bw, 'in[100352]')}},
- {{gm.fi_vdecl(mem_bw, 'out[100352]')}}
-){
- #pragma HLS RESOURCE variable=out core=XPM_MEMORY uram
- #pragma HLS INTERFACE bram port=in
-
- {{gm.f_vdecl_arr(internal_bw,internal_bw, 'out_temp', op_loop_factor)}};
- #pragma HLS ARRAY_PARTITION variable=out_temp
-
-{{gm.ident(1)}}//Clear initial values of output buffers
-{% call gm.forloop(1, 'i', COO*OW*OH*iters_itrm) %}
-{{gm.ident(2)}}#pragma HLS pipeline
-{{gm.ident(2)}}out[i] = 0;
-{% endcall %}
-{{gm.ident(1)}}int weight_idx=0;
-{{gm.ident(1)}}//First set of loops: Iterate over filter
-{% call gm.forloop(1, 'fh', FH) %}
-{% call gm.forloop(2, 'fw', FW) %}
-{% call gm.forloop(3, 'cii', CII) %}
-{% call gm.forloop(4, 'coo', COO) %}
-{{gm.ident(5)}}for (int h=0,ih=0;h<{{OH}};h++,ih+={{stride}}) {
-{{gm.ident(6)}}for (int w=0,iw=0;w<{{OW}};w++,iw+={{stride}}) {
-{{gm.ident(7)}}#pragma HLS pipeline II=2
-{{gm.ident(7)}}{{gm.fi_vdecl(ip_bw*ip_loop_factor,'intemp0')}};
-{{gm.ident(7)}}int address=(cii*{{IW*IH}}+(ih+fh)*{{IW}}+iw+fw)*{{iters_input}};
-{{gm.ident(7)}}readInput_{{object.name}}(in,intemp0,address);
-{% call gm.forloop_s(7, 'i', op_loop_factor, 2) %}
-{{gm.ident(8)}}{{gm.fi_vdecl(w_bw*ip_loop_factor,'weight1')}};
-{{gm.ident(8)}}{{gm.fi_vdecl(w_bw*ip_loop_factor,'weight2')}};
-{{gm.ident(8)}}int address = weight_idx+i/2*{{iters_weight}};
-{{gm.ident(8)}}{{object.name}}_read_weight(weight, weight1, weight2, address);
-{{gm.ident(8)}}#pragma HLS unroll factor={{op_unroll_factor}}
-{{gm.ident(8)}}{{object.name}}_compengine_row(out_temp[i],out_temp[i+1],intemp0,weight1,weight2);
-{% endcall %}
-{{gm.ident(7)}}{{gm.fi_vdecl(int_bw*op_loop_factor,'outPackTemp')}};
-{{gm.ident(7)}}address=(coo*{{OW*OH}}+h*{{OW}}+w)*{{iters_itrm}};
-{{gm.ident(7)}}readItrm_{{object.name}}(out, outPackTemp, address);
-{{gm.ident(7)}}{{object.name}}_out_converge(outPackTemp, out_temp, right_shift);
-{{gm.ident(7)}}writeItrm_{{object.name}}(out, outPackTemp, address);
-{{gm.ident(6)}}}
-{{gm.ident(5)}}}
-{{gm.ident(5)}}weight_idx += ({{op_loop_factor}}/2)*{{iters_weight}};
-{% endcall %}
-{% endcall %}
-{% endcall %}
-{% endcall %}
-}
\ No newline at end of file
diff --git a/DRED_codegen/templates/convolution.inc b/DRED_codegen/templates/convolution.inc
deleted file mode 100644
index c1ddfb760961a66874b973d675bf7eb74b1479a2..0000000000000000000000000000000000000000
--- a/DRED_codegen/templates/convolution.inc
+++ /dev/null
@@ -1,761 +0,0 @@
-{% import 'global_macros.inc' as gm %}
-{% import 'aux_funcs.inc' as aux %}
-
-{%- macro gen_red_tree_vdecl(size, dtype) %}
-{%- for level in range(size, 1, -1) %}
-{% if (size % level) == 0 %}
-{{gm.ident(1)}}{{dtype}} inter{{level}}[2][{{level}}];
-{{gm.ident(1)}}#pragma HLS ARRAY_PARTITION variable=inter{{level}} complete dim=1
-{{gm.ident(1)}}#pragma HLS ARRAY_PARTITION variable=inter{{level}} complete dim=2
-{% endif %}
-{%- endfor %}
-{%- endmacro %}
-
-{%- macro gen_red_tree_impl(size, dtype) %}
-{%- for level in range(size, 1, -1) %}
-{%- if (size % level) == 0 %}
-{% if (level != size) %}
-{{gm.ident(1)}}for (int i=0; i<{{level}}; i++) {
-{{gm.ident(2)}}#pragma HLS unroll
-{{gm.ident(2)}}inter{{level}}[0][i]=inter{{level*2}}[0][i*2]+inter{{level*2}}[0][i*2+1];
-{{gm.ident(2)}}inter{{level}}[1][i]=inter{{level*2}}[1][i*2]+inter{{level*2}}[1][i*2+1];
-{{gm.ident(1)}}}
-{%- endif %}
-{%- endif %}
-{%- endfor %}
-{%- endmacro %}
-
-{%- macro handle_preprocess(object, name) -%}
- {%- if name in object.keys() %}
- {%- for key, value in object.iteritems() %}
- {%- if key == name %}
- {%- if value > 0 %}
- {%- endif %}
- {%- endif %}
- {%- endfor %}
- {%- endif %}
-{%- endmacro %}
-
-{%- macro fdecl(object) -%}
-{%- endmacro %}
-
-{%- macro var_decl(object) -%}
-{%- endmacro %}
-
-{% macro fvalid_call(object) %}
-{{gm.ident(1)}}{{object.name}}_test_conv();
-{{gm.ident(1)}}{{object.name}}_test_relu();
-{% endmacro -%}
-
-{# Function Call #}
-{%- macro fcall(object, debug) -%}
-{%- set ip_shift = object.preprocess.ip_shift-%}
-{%- set int_shift = object.preprocess.int_shift-%}
-{%- set w_shift = object.weights.W_shift-%}
-{%- set op_shift = object.preprocess.op_shift -%}
-{%- set b_shift = object.weights.b_shift-%}
-{%- set op_loop_factor = object.op_loop_factor -%}
-{%- set ip_loop_factor = object.ip_loop_factor -%}
-{%- set right_shift = (ip_shift+w_shift-int_shift) -%}
-{%- set right_shift_relu = int_shift - op_shift -%}
-{%- set left_shift_relu = int_shift-b_shift-%}
-{%- if debug -%}
-{{gm.ident(1)}}{{aux.fcall_readW(object)}}
-{% endif %}
-{# decide the size of output depending on type of padding #}
-{{gm.ident(1)}}{{object.name}}_layer(
-{%- if object.weights["padding"] == 'VALID' %}
-{{object.dshape[0]}}, {{object.dshape[1]}}, {{(object.dshape[2]/ip_loop_factor)|round(0,ceil)|int}}, {{object.dshape[0]-(object.wshape[0]-1)}},{{object.dshape[0]-(object.wshape[0]-1)}},
-{%- else %}
-{{object.dshape[0]+(object.wshape[0]-1)}}, {{object.dshape[1]+(object.wshape[0]-1)}}, {{(object.dshape[2]/ip_loop_factor)|round(0,'ceil')|int}}, {{object.dshape[0]}}, {{object.dshape[1]}},
-{%- endif %}
- {{(object.wshape[3]/op_loop_factor)|round(0,'ceil')|int}}, {{object.wshape[0]}}, {{object.wshape[1]}},
-{%- if object.weights["stride"] %} {#TODO: give more flexibility? #}
- {{object.weights["stride"][1]}},
-{%- else %}
- 1,
-{%- endif %}
- {{right_shift}}, w,
-{%- if object.interface.in == 'dram' %}
- a,
-{%- elif object.interface.in == 'onchip' %}
- a_onchip,
-{%- endif %}
-{%- if object.interface.int == 'dram' %}
- c
-{%- elif object.interface.int == 'onchip' %}
- c_onchip
-{%- endif -%}
-);
-{# if activation function is required #}
-{% if object.activation == 'ReLU' %}
-{%- if debug -%}
-{{aux.fcall_readb(object)}}
-{%- endif -%}
-{{gm.ident(1)}}{{object.name}}_bias_relu(
-{%- if object.weights["padding"] == 'VALID' -%}
-0,
-{%- else %}
- {{object.weights.padsize}},
-{%- endif %}
- {{object.dshape[0]}}, {{object.dshape[1]}}, {{(object.wshape[3]/op_loop_factor)|round(0,'ceil')|int}}, {{object.dshape[0]+2*object.weights.padsize}}, {{object.dshape[1]+2*object.weights.padsize}}, {{left_shift_relu}}, {{right_shift_relu}}, w,
-{%- if object.interface.out == 'dram' %}
-a,
-{%- elif object.interface.out == 'onchip' %}
-a_onchip,
-{% endif %}
-{%- if object.interface.int == 'dram' %}
-c
-{%- elif object.interface.int == 'onchip' %}
-c_onchip
-{%- endif -%}
-);
-{%- endif -%}
-{%- endmacro -%}
-
-{# Macro that implement the memory read & unpack function #}
-{# TODO: share instances of read function between layers #}
-{% macro readFunc(layer, name, mem, bw, loop_factor)%}
-{# Number of packs required to read one pack#}
-{%- set iters = (bw*loop_factor/mem)|round(0,'ceil')|int -%}
-int read{{name}}_{{layer}}({{gm.fixed_int(mem)}} *mem,
- {{gm.fixed_int(bw*loop_factor)}} &input,
- int address)
-{
- int index=0;
- //Read requires {{iters}} iterations
-{%for idx in range(0,iters)%}
-{# if this IS NOT the last pack#}
-{%if ((idx+1)*mem < bw*loop_factor) %}
- input.{{gm.pr_range_bound((idx+1)*mem-1,idx*mem)}}=mem[address+index].{{gm.pr_range_bound((mem-1),0)}};
- index ++;
-{# if this IS the last pack#}
-{% else %}
- input.{{gm.pr_range_bound(bw*loop_factor-1,idx*mem)}}=mem[address+index].{{gm.pr_range_bound(bw*loop_factor-idx*mem-1,0)}};
-{%endif%}
-{%endfor%}
- return address;
-}
-{% endmacro %}
-
-{# Macro that implement the memory write & pack function #}
-{# TODO: share instances of read function between layers #}
-{% macro writeFunc(layer, name, mem, bw, loop_factor)%}
-{%- set iters = (bw*loop_factor/mem)|round(0,'ceil')|int -%}
-int write{{name}}_{{layer}}({{gm.fixed_int(mem)}} *mem,
- {{gm.fixed_int(bw*loop_factor)}} output,
- int address)
-{
- int index=0;
- //Write requires {{iters}} iterations
-{%for idx in range(0,iters)%}
-{# if this IS NOT the last pack#}
-{%if ((idx+1)*mem < bw*loop_factor) %}
- mem[address+index].{{gm.pr_range_bound((mem-1),0)}} = output.{{gm.pr_range_bound((idx+1)*mem-1,idx*mem)}};
- index ++;
-{# if this IS the last pack#}
-{% else %}
- mem[address+index].{{gm.pr_range_bound(bw*loop_factor-idx*mem-1,0)}}=output.{{gm.pr_range_bound(bw*loop_factor-1,idx*mem)}};
-{%endif%}
-{%endfor%}
- return address;
-}
-{% endmacro %}
-
-
-{%- macro fimpl(object) -%}
-{# declare a set of internal variables in Jinja2 for convenience#}
-{# parameters related to quantization of data #}
-{%- set ip_bw = object.preprocess.ip_bit_width-%}
-{%- set ip_shift = object.preprocess.ip_shift-%}
-{%- set int_bw = object.preprocess.int_bit_width-%}
-{%- set int_shift = object.preprocess.int_shift-%}
-{%- set op_bw = object.preprocess.op_bit_width-%}
-{%- set op_shift = object.preprocess.op_shift -%}
-{% if object.preprocess.cluster == true %}
-{%- set index_width = object.preprocess.index_width -%}
-{%- set num_clusters = object.preprocess.num_clusters -%}
-{% endif %}
-{%- set w_bw = object.weights.W_width-%}
-{%- set w_shift = object.weights.W_shift-%}
-{%- set b_bw = object.weights.b_width-%}
-{%- set b_shift = object.weights.b_shift-%}
-{# ip_loop_factor has to be the same as the op_loop_factor of the previous layer #}
-{%- set ip_loop_factor = object.ip_loop_factor -%}
-{%- set op_loop_factor = object.op_loop_factor -%}
-{# mem_bw and onchip_bw can be the same, in that case #}
-{%- set mem_bw = object.interface.memory_bit_width -%}
-{%- set onchip_bw = object.interface.onchip_bit_width -%}
-{# internal bw is the "safe" bit width for the result of multiplication #}
-{%- set internal_bw = w_bw+ip_bw -%}
-{# iterations of read/write required for datapacks #}
-{%- set iters_weight = (w_bw*ip_loop_factor*2/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_bias = (ip_bw*ip_loop_factor/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_input = (ip_bw*ip_loop_factor/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_output = (op_bw*op_loop_factor/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_itrm = (int_bw*op_loop_factor/mem_bw)|round(0,'ceil')|int -%}
-
-{# Implement the memory read & unpack function #}
-{# see the overview doc for more detail #}
-{% if object.preprocess.cluster == true%}
-int {{object.name}}_read_weight({{gm.fixed_int(mem_bw)}} *mem,
- {{gm.fixed_int(index_width*ip_loop_factor)}} &weight1,
- {{gm.fixed_int(index_width*ip_loop_factor)}} &weight2,
- int address)
-{
- int index=0;
- //Input loop factor is {{ip_loop_factor}}
- //Length of pack is {{index_width*ip_loop_factor}}
- //Read requires {{iters_weight}} iterations
-{%for idx in range(0,iters_weight)%}
-{# if this is not the last pack#}
-{%if ((idx+1)*(mem_bw/2) < index_width*ip_loop_factor) %}
- weight1.{{gm.pr_range_bound(((idx+1)*(mem_bw/2)-1)|int,(idx*mem_bw/2)|int)}}=mem[address+index].{{gm.pr_range_bound((mem_bw/2-1)|int,0)}};
- weight2.{{gm.pr_range_bound(((idx+1)*(mem_bw/2)-1)|int,(idx*mem_bw/2)|int)}}=mem[address+index].{{gm.pr_range_bound(mem_bw-1,(mem_bw/2)|int)}};
- index+=1;
-{# if this is the last pack#}
-{% else %}
- weight1.{{gm.pr_range_bound(index_width*ip_loop_factor-1,(idx*mem_bw/2)|int)}}=mem[address+index].{{gm.pr_range_bound(index_width*ip_loop_factor-1-(idx*mem_bw/2)|int,0)}};
- weight2.{{gm.pr_range_bound(index_width*ip_loop_factor-1,(idx*mem_bw/2)|int)}}=mem[address+index].{{gm.pr_range_bound(index_width*ip_loop_factor-1-((idx-1)*mem_bw/2)|int,(mem_bw/2)|int)}};
-{%endif%}
-{%endfor%}
- return address;
-}
-{% else %}
-int {{object.name}}_read_weight({{gm.fixed_int(mem_bw)}} *mem,
- {{gm.fixed_int(w_bw*ip_loop_factor)}} &weight1,
- {{gm.fixed_int(w_bw*ip_loop_factor)}} &weight2,
- int address)
-{
- int index=0;
- //Input loop factor is {{ip_loop_factor}}
- //Length of pack is {{w_bw*ip_loop_factor}}
- //Read requires {{iters_weight}} iterations
-{%for idx in range(0,iters_weight)%}
-{# if this is not the last pack#}
-{%if ((idx+1)*(mem_bw/2) < w_bw*ip_loop_factor) %}
- weight1.{{gm.pr_range_bound(((idx+1)*(mem_bw/2)-1)|int,(idx*mem_bw/2)|int)}}=mem[address+index].{{gm.pr_range_bound((mem_bw/2-1)|int,0)}};
- weight2.{{gm.pr_range_bound(((idx+1)*(mem_bw/2)-1)|int,(idx*mem_bw/2)|int)}}=mem[address+index].{{gm.pr_range_bound(mem_bw-1,(mem_bw/2)|int)}};
- index+=1;
-{# if this is the last pack#}
-{% else %}
- weight1.{{gm.pr_range_bound(w_bw*ip_loop_factor-1,(idx*mem_bw/2)|int)}}=mem[address+index].{{gm.pr_range_bound(w_bw*ip_loop_factor-1-(idx*mem_bw/2)|int,0)}};
- weight2.{{gm.pr_range_bound(w_bw*ip_loop_factor-1,(idx*mem_bw/2)|int)}}=mem[address+index].{{gm.pr_range_bound(w_bw*ip_loop_factor-1-((idx-1)*mem_bw/2)|int,(mem_bw/2)|int)}};
-{%endif%}
-{%endfor%}
- return address;
-}
-{% endif %}
-
-{# Declare Read functions for this layer#}
-{{readFunc(object.name, "Input", mem_bw, ip_bw, ip_loop_factor)}}
-{{readFunc(object.name,"Itrm", mem_bw, int_bw, op_loop_factor)}}
-{{readFunc(object.name,"Bias", mem_bw, b_bw, op_loop_factor)}}
-{{writeFunc(object.name,"Itrm", mem_bw, int_bw, op_loop_factor)}}
-{{writeFunc(object.name,"Output", mem_bw, op_bw, op_loop_factor)}}
-
-
-{% if object.preprocess.cluster == true %}
-{{gm.fixed(w_bw, w_bw)}} weightTable({{gm.fui_vdecl(index_width, 'index')}})
-{
-{{gm.ident(1)}}const {{gm.f_vdecl_arr(w_bw, w_bw, 'table', num_clusters)}} =
-{{gm.ident(1)}}{
-{%- for weight in object.cluster_values -%}
-{# this if is just to avoid that ',' for the last iteration #}
-{%- if loop.index == num_clusters -%}
-{{weight}}
-{% else %}
-{{weight}},
-{%- endif -%}
-{% endfor %}
-{{gm.ident(1)}}};
-{{gm.ident(1)}}return table[index];
-}
-
-void {{object.name}}_decode_weight({{gm.fi_vdecl_arr(w_bw, 'weightArray', ip_loop_factor)}}, {{gm.fi_vdecl(ip_loop_factor*index_width, 'indexPack')}})
-{
-{{gm.ident(1)}}#pragma HLS ARRAYPARTITION variable = weightArray complete
-{% call gm.forloop(1, 'i', ip_loop_factor) %}
-{{gm.ident(2)}}#pragma HLS unroll
-{{gm.ident(2)}}ap_uint<{{index_width}}> indexTmp;
-{{gm.ident(2)}}indexTmp.{{gm.pr_range(index_width)}}=indexPack.{{gm.pr_range_idx(index_width, 'i')}};
-{{gm.ident(2)}}weightArray[i]=weightTable(indexTmp);
-{%- endcall -%}
-}
-{% endif %}
-
-{# determine whether multiplication can be packed#}
-{%- if ((ip_bw > 8) or (w_bw > 8) or (internal_bw > 16)) -%}
-{# currently we only allow 8bit, might expand in the future#}
-{# simple compute function when we cannot pack 2 multiplications #}
-void {{object.name}}_pack_dsp (
- {{gm.f_vdecl(internal_bw, internal_bw,'&rst1')}},
- {{gm.f_vdecl(internal_bw, internal_bw,'&rst2')}},
- {{gm.fi_vdecl(w_bw, 'a')}},
- {{gm.fi_vdecl(w_bw, 'b')}},
- {{gm.fi_vdecl(ip_bw, 'c')}}
-) {
- rst1 = a * c;
- rst2 = b * c;
-}
-{%- else -%}
-void {{object.name}}_pack_dsp(
- {{gm.f_vdecl(internal_bw, internal_bw,'&rst1')}},
- {{gm.f_vdecl(internal_bw, internal_bw,'&rst2')}},
- {{gm.fi_vdecl(w_bw, 'a')}},
- {{gm.fi_vdecl(w_bw, 'b')}},
- {{gm.fi_vdecl(ip_bw, 'c')}}
-) {
- {{gm.fi_vdecl(18+w_bw, 'ai')}} = 0;
- {{gm.fi_vdecl(w_bw, 'bi')}};
- {{gm.fi_vdecl(ip_bw, 'ci')}};
- {{gm.fi_vdecl(45, 'rst')}};
- {{gm.fui_vdecl(19, 'cst')}};
-
- #pragma HLS RESOURCE variable=rst core=DSP48
- #pragma HLS pipeline II=1
- ai.{{gm.pr_range(w_bw)}}=a.{{gm.pr_range(w_bw)}};
- bi.{{gm.pr_range(w_bw)}}=b.{{gm.pr_range(w_bw)}};
- ci.{{gm.pr_range(ip_bw)}}=c.{{gm.pr_range(ip_bw)}};
-
- if ( (ci!=0) && (bi!=0) && (ci.sign() != bi.sign())) {
- cst=0x40000;
- } else {
- cst=0;
- }
- rst = ( (ai<<18) + bi) * ci + cst;
- rst1.{{gm.pr_range(internal_bw)}}=rst.{{gm.pr_range_bound(18+internal_bw-1,18)}};
- rst2.{{gm.pr_range(internal_bw)}}=rst.{{gm.pr_range(internal_bw)}};
-}
-{% endif %}
-
-void {{object.name}}_out_converge({{gm.fi_vdecl(op_loop_factor*int_bw, '&outpack')}}, {{gm.f_vdecl_arr(internal_bw,internal_bw,'out_temp',op_loop_factor)}}, int right_shift) {
- #pragma HLS pipeline
- #pragma HLS ARRAY_PARTITION variable=out_temp complete dim=1
- {{gm.f_vdecl_arr(int_bw, int_bw,'temp_1',op_loop_factor)}};
- {{gm.f_vdecl_arr(internal_bw, internal_bw,'temp_2',op_loop_factor)}};
- #pragma HLS ARRAY_PARTITION variable=temp_1 complete dim=1
- #pragma HLS ARRAY_PARTITION variable=temp_2 complete dim=1
-{% call gm.forloop(1,'i',op_loop_factor) %}
-{{gm.ident(2)}}#pragma HLS unroll factor=64
-{{gm.ident(2)}}temp_1[i].{{gm.pr_range(int_bw)}}=outpack.{{gm.pr_range_idx(int_bw, 'i')}};
-{{gm.ident(2)}}temp_2[i].{{gm.pr_range(internal_bw)}}=out_temp[i].{{gm.pr_range(internal_bw)}};
-{{gm.ident(2)}}temp_1[i]=temp_1[i] + (temp_2[i] >> right_shift);
-{%- endcall -%}
-{% call gm.forloop(1,'i',op_loop_factor) %}
-{{gm.ident(2)}}#pragma HLS unroll factor=64
-{{gm.ident(2)}}outpack.{{gm.pr_range_idx(int_bw, 'i')}}=temp_1[i].{{gm.pr_range(int_bw)}};
-{%- endcall -%}
-}
-
-void {{object.name}}_compengine_row(
- {{gm.f_vdecl(internal_bw,internal_bw,'&out1')}},
- {{gm.f_vdecl(internal_bw,internal_bw,'&out2')}},
- {{gm.fi_vdecl(ip_loop_factor*ip_bw, 'InPack')}},
- {% if object.preprocess.cluster == true -%}
- {{gm.fi_vdecl(index_width*ip_loop_factor, 'weightPack1')}},
- {{gm.fi_vdecl(index_width*ip_loop_factor, 'weightPack2')}}
- {%- else -%}
- {{gm.fi_vdecl(w_bw*ip_loop_factor, 'weightPack1')}},
- {{gm.fi_vdecl(w_bw*ip_loop_factor, 'weightPack2')}}
- {%- endif -%}
-) {
- {% if object.preprocess.cluster == true %}
- {{gm.fi_vdecl_arr(w_bw, 'weight1', ip_loop_factor)}};
- #pragma HLS ARRAY_PARTITION variable=weight1 complete dim=1
- {{gm.fi_vdecl_arr(w_bw, 'weight2', ip_loop_factor)}};
- #pragma HLS ARRAY_PARTITION variable=weight2 complete dim=1
- {% else %}
- {{gm.fi_vdecl_arr(index_width, 'weight1', ip_loop_factor)}};
- #pragma HLS ARRAY_PARTITION variable=weight1 complete dim=1
- {{gm.fi_vdecl_arr(index_width, 'weight2', ip_loop_factor)}};
- #pragma HLS ARRAY_PARTITION variable=weight2 complete dim=1
- {% endif %}
- {{gm.fi_vdecl_arr(ip_bw, 'in', ip_loop_factor)}};
- #pragma HLS ARRAY_PARTITION variable=in complete dim=1
- {{gen_red_tree_vdecl(ip_loop_factor, gm.fixed(internal_bw,internal_bw))}}
- #pragma HLS pipeline II=2
-{% call gm.forloop(1, 'i', ip_loop_factor) %}
-{{gm.ident(2)}}in[i].{{gm.pr_range(ip_bw)}}=InPack.{{gm.pr_range_idx(ip_bw,'i')}};
-{# if cluster is disabled #}
-{% if object.preprocess.cluster == false %}
-{{gm.ident(2)}}weight1[i].{{gm.pr_range(w_bw)}}=weightPack1.{{gm.pr_range_idx(w_bw,'i')}};
-{{gm.ident(2)}}weight2[i].{{gm.pr_range(w_bw)}}=weightPack2.{{gm.pr_range_idx(w_bw,'i')}};
-{% endif %}
-{%- endcall %}
-{# if cluster is enabled #}
-{%- if object.preprocess.cluster == true -%}
-{{gm.ident(1)}}//Decode clustered input
-{{gm.ident(1)}}{{object.name}}_decode_weight(weight1, weightPack1);
-{{gm.ident(1)}}{{object.name}}_decode_weight(weight2, weightPack2);
-{% endif %}
-{%- call gm.forloop(1, 'i', ip_loop_factor) %}
-{{gm.ident(2)}}#pragma HLS unroll
-{{gm.ident(2)}}{{object.name}}_pack_dsp(inter{{ip_loop_factor}}[0][i],inter{{ip_loop_factor}}[1][i],weight1[i],weight2[i],in[i]);
-{%- endcall %}
-{{gen_red_tree_impl(ip_loop_factor, gm.fixed(internal_bw,internal_bw))}}
- out1=inter2[0][0]+inter2[0][1];
- out2=inter2[1][0]+inter2[1][1];
-}
-
-{{gm.fixed_int(op_bw*op_loop_factor)}} {{object.name}}_bias_relu(
- {{gm.f_vdecl_arr(int_bw, int_bw,'out1',op_loop_factor)}},
- {{gm.fi_vdecl(b_bw*op_loop_factor, 'bias')}},
- int left_shift,
- int right_shift
-){
- {{gm.f_vdecl_arr(op_bw, op_bw, 'in1', op_loop_factor)}};
- {{gm.fi_vdecl(op_bw*op_loop_factor, 'temp')}};
- #pragma HLS inline off
- #pragma HLS array_partition variable=out1 dim=1
-{% call gm.forloop(1, 'j', op_loop_factor) %}
-{{gm.ident(2)}}#pragma HLS unroll
-{{gm.ident(2)}}{{gm.f_vdecl(b_bw, b_bw ,'btemp')}};
-{{gm.ident(2)}}{{gm.f_vdecl(int_bw, int_bw,'otemp')}};
-{{gm.ident(2)}}btemp.{{gm.pr_range(b_bw)}} = bias.{{gm.pr_range_idx(b_bw,'j')}};
-{{gm.ident(2)}}otemp=(out1[j] + (({{gm.fixed(int_bw, int_bw)}})btemp << left_shift))>>right_shift;
-{{gm.ident(2)}}if(otemp>0)
-{{gm.ident(3)}}in1[j]=otemp;
-{{gm.ident(2)}}else
-{{gm.ident(3)}}in1[j]=0.0;
-{{gm.ident(2)}}temp.{{gm.pr_range_idx(op_bw,'j')}}=in1[j].{{gm.pr_range(op_bw)}};
-{%- endcall -%}
- return temp;
-}
-
-void {{object.name}}_layer(
- int IW, //image width with padding
- int IH, //image height with padding
- int CII,//image channels
- int OW, //output width
- int OH, //output height
- int COO, //output channels
- int FW, //filter width
- int FH, //filter height
- int stride,
- int right_shift,
- {{gm.fi_vdecl(mem_bw, '*weight')}},
- {% if object.interface.in == 'dram' %}
- {{gm.fi_vdecl(mem_bw, '*in')}},
- {% elif object.interface.in == 'onchip' %}
- {{gm.fi_vdecl_arr(onchip_bw, 'in', 50176)}},
- {% endif %}
- {% if object.interface.int == 'dram' %}
- {{gm.fi_vdecl(mem_bw, '*out')}}
- {% elif object.interface.int == 'onchip' %}
- {{gm.fi_vdecl_arr(onchip_bw, 'out', 50176)}}
- {% endif %}
-){
- #pragma HLS RESOURCE variable=out core=XPM_MEMORY uram
- #pragma HLS INTERFACE bram port=in
-
- {{gm.f_vdecl_arr(internal_bw,internal_bw, 'out_temp', op_loop_factor)}};
- #pragma HLS ARRAY_PARTITION variable=out_temp
-
-{{gm.ident(1)}}//Clear initial values of output buffers
-{% if (object.interface.int == 'dram'): %}
-{% call gm.forloop(1, 'i', 'COO*OW*OH*%d'|format(iters_itrm)) %}
-{{gm.ident(2)}}#pragma HLS pipeline
-{{gm.ident(2)}}out[i] = 0;
-{% endcall %}
-{% elif (object.interface.int == 'onchip'): %}
-{% call gm.forloop(1, 'i', 'COO*OW*OH*%d'|format(iters_itrm)) %}
-{{gm.ident(2)}}#pragma HLS pipeline
-{{gm.ident(2)}}out[i] = 0;
-{% endcall %}
-{% endif %}
-{{gm.ident(1)}}int weight_idx=0;
-{{gm.ident(1)}}//First set of loops: Iterate over filter
-{% call gm.forloop(1, 'fh', 'FH') %}
-{% call gm.forloop(2, 'fw', 'FW') %}
-{% call gm.forloop(3, 'cii', 'CII') %}
-{% call gm.forloop(4, 'coo', 'COO') %}
-{{gm.ident(5)}}#pragma HLS dataflow
-{{gm.ident(5)}}for (int h=0,ih=0;h<OH;h++,ih+=stride) {
-{{gm.ident(6)}}for (int w=0,iw=0;w<OW;w++,iw+=stride) {
-{{gm.ident(7)}}#pragma HLS pipeline II=2
-{{gm.ident(7)}}{{gm.fi_vdecl(ip_bw*ip_loop_factor,'intemp0')}};
-{% if object.interface.in == 'dram' %}
-{{gm.ident(7)}}int address=(cii*IW*IH+(ih+fh)*IW+iw+fw)*{{iters_input}};
-{{gm.ident(7)}}readInput_{{object.name}}(in,intemp0,address);
-{% elif object.interface.in == 'onchip' %}
-{{gm.ident(7)}}int address=(cii*IW*IH+(ih+fh)*IW+iw+fw)*{{iters_input}};
-{{gm.ident(7)}}readInput_{{object.name}}(in,intemp0,address);
-{% endif %}
-{% call gm.forloop_s(7, 'i', op_loop_factor, 2) %}
-{% if object.preprocess.cluster == true %}
-{{gm.ident(8)}}{{gm.fi_vdecl(index_width*ip_loop_factor,'weight1')}};
-{{gm.ident(8)}}{{gm.fi_vdecl(index_width*ip_loop_factor,'weight2')}};
-{% else %}
-{{gm.ident(8)}}{{gm.fi_vdecl(w_bw*ip_loop_factor,'weight1')}};
-{{gm.ident(8)}}{{gm.fi_vdecl(w_bw*ip_loop_factor,'weight2')}};
-{% endif%}
-{{gm.ident(8)}}int address = weight_idx+i/2*{{iters_weight}};
-{{gm.ident(8)}}{{object.name}}_read_weight(weight, weight1, weight2, address);
-{{gm.ident(8)}}#pragma HLS unroll factor=32
-{{gm.ident(8)}}{{object.name}}_compengine_row(out_temp[i],out_temp[i+1],intemp0,weight1,weight2);
-{% endcall %}
-{{gm.ident(7)}}{{gm.fi_vdecl(int_bw*op_loop_factor,'outPackTemp')}};
-{% if object.interface.int == 'dram' %}
-{{gm.ident(7)}}address=(coo*OW*OH+h*OW+w)*{{iters_itrm}};
-{{gm.ident(7)}}readItrm_{{object.name}}(out, outPackTemp, address);
-{{gm.ident(7)}}{{object.name}}_out_converge(outPackTemp, out_temp, right_shift);
-{{gm.ident(7)}}writeItrm_{{object.name}}(out, outPackTemp, address);
-{% elif object.interface.int == 'onchip' %}
-{{gm.ident(7)}}address=(coo*OW*OH+h*OW+w)*{{iters_itrm}};
-{{gm.ident(7)}}readItrm_{{object.name}}(out, outPackTemp, address);
-{{gm.ident(7)}}{{object.name}}_out_converge(outPackTemp, out_temp, right_shift);
-{{gm.ident(7)}}writeItrm_{{object.name}}(out, outPackTemp, address);
-{% endif %}
-{{gm.ident(6)}}}
-{{gm.ident(5)}}}
-{{gm.ident(5)}}weight_idx += ({{op_loop_factor}}/2)*{{iters_weight}};
-{% endcall %}
-{% endcall %}
-{% endcall %}
-{% endcall %}
-}
-
-void {{object.name}}_bias_relu(
- int padsize,
- int OH, //size of output of conv_layer
- int OW, //size of output of conv_layer
- int COO, //channels
- int IH, //size of output of relu layer
- int IW, //size of output of relu layer
- int left_shift,
- int right_shift,
- {{gm.fi_vdecl(mem_bw, '*weight')}},
- {% if object.interface.out == 'dram' %}
- {{gm.fi_vdecl(mem_bw, '*in')}},
- {% elif object.interface.out == 'onchip' %}
- {{gm.fi_vdecl_arr(onchip_bw, 'in', 50176)}},
- {% endif %}
- {% if object.interface.int == 'dram' %}
- {{gm.fi_vdecl(mem_bw, '*out')}}
- {% elif object.interface.int == 'onchip' %}
- {{gm.fi_vdecl_arr(onchip_bw, 'out', 50176)}}
- {% endif %}
-) {
- {{gm.fi_vdecl(op_loop_factor*int_bw, 'temp')}};
- {{gm.fi_vdecl(op_loop_factor*b_bw, 'bias')}};
- int weight_idx = 0;
- int address = 0;
- {{gm.fi_vdecl(op_loop_factor*op_bw,"outPackTemp")}};
-{% call gm.forloop(1,'c', 'COO') %}
-{{gm.ident(2)}}readBias_{{object.name}}(weight, bias, weight_idx*{{iters_bias}});
-{{gm.ident(2)}}weight_idx ++;
-{% call gm.forloop(2, 'i', 'padsize*IW') %}
-{{gm.ident(3)}}#pragma HLS pipeline
-{% if object.interface.out == 'dram' %}
-{{gm.ident(3)}}address = (c*IH*IW+i)*{{iters_output}};
-{{gm.ident(3)}}writeOutput_{{object.name}}(in, 0, address);
-{% elif object.interface.out == 'onchip' %}
-{{gm.ident(3)}}address = (c*IH*IW+i)*{{iters_output}};
-{{gm.ident(3)}}writeOutput_{{object.name}}(in, 0, address);
-{% endif %}
-{% endcall %}
-{% call gm.forloop(2, 'h', 'OH') %}
-{{gm.ident(3)}}// pad zeros in front of each line
-{% call gm.forloop(3, 'i', 'padsize') %}
-{{gm.ident(4)}}#pragma HLS pipeline
-{% if object.interface.out == 'dram' %}
-{{gm.ident(4)}}address = (c*IH*IW+(padsize+h)*IW+i)*{{iters_output}};
-{{gm.ident(4)}}writeOutput_{{object.name}}(in, 0, address);
-{% elif object.interface.out == 'onchip' %}
-{{gm.ident(4)}}address = (c*IH*IW+(padsize+h)*IW+i)*{{iters_output}};
-{{gm.ident(4)}}writeOutput_{{object.name}}(in, 0, address);
-{% endif %}
-{% endcall %}
-{{gm.ident(3)}}// perform relu & bias
-{% call gm.forloop(3, 'w', 'OW') %}
-{{gm.ident(4)}}#pragma HLS pipeline
-{{gm.ident(4)}}{{gm.f_vdecl_arr(int_bw, int_bw, 'out_temp', op_loop_factor)}};
-{{gm.ident(4)}}#pragma HLS array_partition variable=out_temp
-{% if object.interface.int == 'dram' %}
-{{gm.ident(4)}}address = (c*OW*OH+h*OW+w)*{{iters_itrm}};
-{{gm.ident(4)}}readItrm_{{object.name}}(out, temp, address);
-{% elif object.interface.int == 'onchip' %}
-{{gm.ident(4)}}address = (c*OW*OH+h*OW+w)*{{iters_itrm}};
-{{gm.ident(4)}}readItrm_{{object.name}}(out, temp, address);
-{% endif %}
-{% call gm.forloop(4,'j',op_loop_factor) %}
-{{gm.ident(5)}}#pragma HLS unroll
-{{gm.ident(5)}}out_temp[j].{{gm.pr_range(int_bw)}}=temp.{{gm.pr_range_idx(int_bw,'j')}};
-{% endcall %}
-{% if object.interface.out == 'dram' %}
-{{gm.ident(4)}}address = (c*IH*IW+(padsize+h)*IW+w+padsize)*{{iters_output}};
-{{gm.ident(4)}}outPackTemp = {{object.name}}_bias_relu(out_temp,bias,left_shift,right_shift);
-{{gm.ident(4)}}writeOutput_{{object.name}}(in, outPackTemp, address);
-{% elif object.interface.out == 'onchip' %}
-{{gm.ident(4)}}address = (c*IH*IW+(padsize+h)*IW+w+padsize)*{{iters_output}};
-{{gm.ident(4)}}outPackTemp = {{object.name}}_bias_relu(out_temp,bias,left_shift,right_shift);
-{{gm.ident(4)}}writeOutput_{{object.name}}(in, outPackTemp, address);
-{% endif %}
-{% endcall %}
-{{gm.ident(3)}}// pad zeros at end of each line
-{% call gm.forloop(3, 'i', 'padsize') %}
-{{gm.ident(4)}}#pragma HLS pipeline
-{% if object.interface.out == 'dram' %}
-{{gm.ident(4)}}address = (c*IH*IW+(padsize+h)*IW+OW+padsize+i)*{{iters_output}};
-{{gm.ident(4)}}writeOutput_{{object.name}}(in, 0, address);
-{% elif object.interface.out == 'onchip' %}
-{{gm.ident(4)}}address = (c*IH*IW+(padsize+h)*IW+OW+padsize+i)*{{iters_output}};
-{{gm.ident(4)}}writeOutput_{{object.name}}(in, 0, address);
-{% endif %}
-{% endcall %}
-{% endcall %}
-{{gm.ident(2)}}// pad trailing zero rows
-{% call gm.forloop(2, 'i', 'padsize*IW') %}
-{{gm.ident(3)}}#pragma HLS pipeline
-{% if object.interface.out == 'dram' %}
-{{gm.ident(3)}}address = (c*IH*IW+(padsize+OH)*IW+i);
-{{gm.ident(3)}}writeOutput_{{object.name}}(in, 0, address);
-{% elif object.interface.out == 'onchip' %}
-{{gm.ident(3)}}address = (c*IH*IW+(padsize+OH)*IW+i);
-{{gm.ident(3)}}writeOutput_{{object.name}}(in, 0, address);
-{% endif %}
-{% endcall %}
-{% endcall %}
-}
-{%- endmacro %}
-
-{%- macro fvalid(object) -%}
-{%- set H = object.dshape[0] -%}
-{%- set W = object.dshape[1] -%}
-{%- set FH = object.wshape[0] -%}
-{%- set FW = object.wshape[1] -%}
-{%- set IC = object.wshape[2] -%}
-{%- set OC = object.wshape[3] -%}
-{%- set padsize = ((FH-1)/2)|int -%}
-{%- set IH = H+2*padsize -%}
-{%- set IW = W+2*padsize -%}
-{%- set ip_bw = object.preprocess.ip_bit_width-%}
-{%- set int_bw = object.preprocess.int_bit_width-%}
-{%- set op_bw = object.preprocess.op_bit_width-%}
-{%- set w_bw = object.weights.W_width-%}
-{%- set b_bw = object.weights.b_width-%}
-{%- set ip_loop_factor = object.ip_loop_factor -%}
-{%- set op_loop_factor = object.op_loop_factor -%}
-{%- set mem_bw = object.interface.memory_bit_width -%}
-{%- set onchip_bw = object.interface.onchip_bit_width -%}
-{%- set iters_weight = (w_bw*ip_loop_factor*2/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_bias = (ip_bw*ip_loop_factor/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_input = (ip_bw*ip_loop_factor/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_output = (op_bw*op_loop_factor/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_itrm = (int_bw*op_loop_factor/mem_bw)|round(0,'ceil')|int -%}
-void {{object.name}}_test_conv() {
- srand(17);
- //test framework
- int input_buf[{{IH}}*{{IW}}*{{IC}}];
- int weight_buf[{{FH}}*{{FW}}*{{IC}}*{{OC}}];
- int output_buf[{{H}}*{{W}}*{{OC}}] = {0};
-{% call gm.forloop(1, 'i', IC) %}
-{% call gm.forloop(2, 'j', IH) %}
-{% call gm.forloop(3, 'k', IW) %}
-{{gm.ident(4)}}if(!(k < {{padsize}} || k >= {{W+padsize}} || j < {{padsize}} || j >= {{H+padsize}})){
-{{gm.ident(5)}}input_buf[i*{{IH}}*{{IW}}+j*{{IW}}+k] = rand()%2;
-{{gm.ident(4)}}} else {
-{{gm.ident(5)}}input_buf[i*{{IH}}*{{IW}}+j*{{IW}}+k] = 0;
-{{gm.ident(4)}}}
-{% endcall %}
-{% endcall %}
-{% endcall %}
-
-{% call gm.forloop(1, 'i', (IC/ip_loop_factor)|round(0,'ceil')|int) %}
-{% call gm.forloop(2, 'j', IH) %}
-{% call gm.forloop(3, 'k', IW) %}
-{{gm.ident(4)}}{{gm.fi_vdecl(512, 'temp_a')}} = 0;
-{% call gm.forloop(4, 'ii', ip_loop_factor) %}
-{{gm.ident(5)}}if(ii*{{ip_loop_factor}}+i < {{IC}}){
-{{gm.ident(6)}}temp_a.{{gm.pr_range_idx(w_bw,'ii')}} = input_buf[(i*{{ip_loop_factor}}+ii)*{{IH}}*{{IW}}+j*{{IW}}+k];
-{{gm.ident(5)}}} else {
-{{gm.ident(6)}}temp_a.{{gm.pr_range_idx(w_bw,'ii')}} = 0;
-{{gm.ident(5)}}}
-{% endcall %}
-{{gm.ident(4)}}a[i*{{IH}}*{{IW}}+j*{{IW}}+k]=temp_a;
-{% endcall %}
-{% endcall %}
-{% endcall %}
-
-{% call gm.forloop(1, 'i', FH*FW*IC*OC) %}
-{{gm.ident(2)}}weight_buf[i] = rand()%2;
-{% endcall %}
-pack_weights<int>({{FH}}, {{FW}}, {{IC}}, {{OC}}, weight_buf, w);
-
-{% call gm.forloop(1, 'i', 50176) %}
-{{gm.ident(2)}}c[i] = 0;
-{% endcall %}
-
-{{gm.ident(1)}}conv_2d(output_buf, weight_buf, input_buf, {{FH}}, {{FW}}, {{IH}}, {{IW}}, {{IC}}, {{H}}, {{W}}, {{OC}}, 1);
-{{gm.ident(1)}}{{object.name}}_layer( {{IW}}, {{IH}}, {{(IC/ip_loop_factor)|round(0,'ceil')|int}} , {{W}}, {{H}}, {{(OC/op_loop_factor)|round(0,'ceil')|int}}, {{FW}}, {{FH}}, 1, 0, w, a, c);
-{% call gm.forloop(1, 'i', 1) %}
-{% call gm.forloop(2, 'ii', op_loop_factor) %}
-{% call gm.forloop(3, 'j', H) %}
-{% call gm.forloop(4, 'k', W) %}
-{{gm.ident(5)}}{{gm.fi_vdecl(512, 'temp_c')}} = c[(i*{{H}}*{{W}}+j*{{W}}+k)*{{iters_itrm}}];
-{{gm.ident(5)}}{{gm.fi_vdecl(16, 'out_layer')}} = temp_c.{{gm.pr_range_idx(16,'ii')}};
-{{gm.ident(5)}}int out_golden = output_buf[(i*{{op_loop_factor}}+ii)*{{H}}*{{W}}+j*{{W}}+k];
-{{gm.ident(5)}}if (out_layer != out_golden) {
-{{gm.ident(5)}}cout << "[" << (i*{{op_loop_factor}}+ii) << "][" << j << "][" << k << "]";
-{{gm.ident(5)}}cout << "Diff: Layer - " << out_layer << "Golden - " << out_golden << endl;
-{{gm.ident(5)}}}
-{%- endcall -%}
-{{gm.ident(3)}}//cout << endl;
-{%- endcall -%}
-{{gm.ident(2)}}//cout << endl;
-{%- endcall -%}
-{%- endcall -%}
-}
-
-
-void {{object.name}}_test_relu() {
- //test framework
- int a_buf[D1][D2][D3];
- const int padsize = {{((object.wshape[0]-1)//2)}}; //for 3x3 kernel
-{% call gm.forloop(1, 'i', 'D1') %}
-{% call gm.forloop(2, 'j', 'D2') %}
-{% call gm.forloop(3, 'k', 'D3') %}
-{{gm.ident(4)}}a_buf[i][j][k] = (i+j-k);
-{% endcall %}
-{% endcall %}
-{% endcall %}
-
-{% call gm.forloop(1, 'i', 'D1') %}
-{% call gm.forloop(2, 'j', 'D2') %}
-{% call gm.forloop(3, 'k', 'D3/64') %}
-{{gm.ident(4)}}{{gm.fi_vdecl(1024, 'temp_c')}} = 0;
-{% call gm.forloop(4, 'ii', 64) %}
-{{gm.ident(5)}}temp_c.{{gm.pr_range_idx(16,'ii')}} = a_buf[i][j][k*64+ii];
-{% endcall %}
-{{gm.ident(4)}}c[k*(D1*D2)+(j)*D1+i]=temp_c;
-{% endcall %}
-{% endcall %}
-{% endcall %}
- {{gm.fi_vdecl(512, 'tmp')}};
-{% call gm.forloop(1, 'i', 64) %}
-{{gm.ident(2)}}tmp.{{gm.pr_range_idx(8, 'i')}} = 1;
-{% endcall %}
-{% call gm.forloop(1, 'i', 'F*F*D4') %}
-{{gm.ident(2)}}w[i] = tmp;
-{% endcall %}
-{% call gm.forloop(1, 'i', 50176) %}
-{{gm.ident(2)}}a[i] = 0;
-{% endcall %}
- //{{object.name}}_bias_relu(padsize, D1, D2, D4/64, D1+2*padsize, D2+2*padsize, w, a, c);
-
-{% call gm.forloop(1, 'i', 'D1+padsize*2') %}
-{% call gm.forloop(2, 'j', 'D2+padsize*2') %}
-{% call gm.forloop(3, 'k', 'D4/64') %}
-{{gm.ident(4)}}{{gm.fi_vdecl(512, 'temp_a')}} = a[k*(D1+2*padsize)*(D2+2*padsize)+j*(D1+2*padsize)+i];
-{{gm.ident(4)}}cout << "[" << i << "][" << j << "][" << k << "]";
-{% call gm.forloop(4, 'ii', 64) %}
-{{gm.ident(5)}}{{gm.fi_vdecl(8, 't')}} = temp_a.{{gm.pr_range_idx(8,'ii')}};
-{{gm.ident(5)}}cout << t << ",";
-{% endcall %}
-{{gm.ident(4)}}cout << endl;
-{% endcall %}
-{% endcall %}
-{% endcall %}
-}
-{%- endmacro %}
\ No newline at end of file
diff --git a/DRED_codegen/templates/fc.inc b/DRED_codegen/templates/fc.inc
deleted file mode 100644
index b294b2f678025f87e04b5fa0e3870a026af6b9bb..0000000000000000000000000000000000000000
--- a/DRED_codegen/templates/fc.inc
+++ /dev/null
@@ -1,781 +0,0 @@
-{# fc.inc #}
-{# Jinja2 template for HLS synthesizable C++ code #}
-{# Inspirit IoT, Inc. 2018 #}
-{# Contributers: #}
-{# Mang Yu #}
-{# Kyle Rupnow #}
-{# Xinheng Liu #}
-
-{% import "global_macros.inc" as gm %}
-{% set ident = gm.ident %}
-{% import 'aux_funcs.inc' as aux %}
-
-{# Description: generate intermediate buffer for adder tree #}
-{# ------ #}
-{# Params: #}
-{# size -- number of items for the input to adder tree #}
-{# dtype -- datatype of the input #}
-{%- macro gen_red_tree_vdecl(size, dtype) %}
-{%- for level in range(size, 1, -1) %}
-{% if (size % level) == 0 %}
-{{gm.ident(1)}}{{dtype}} inter{{level}}[2][{{level}}];
-{{gm.ident(1)}}#pragma HLS ARRAY_PARTITION variable=inter{{level}} complete dim=1
-{{gm.ident(1)}}#pragma HLS ARRAY_PARTITION variable=inter{{level}} complete dim=2
-{% endif %}
-{%- endfor %}
-{%- endmacro %}
-
-{# Description: generate cascaded adding for adder tree #}
-{# ------ #}
-{# Params: #}
-{# size -- number of items for the input to adder tree #}
-{# dtype -- datatype of the input #}
-{%- macro gen_red_tree_impl(size, dtype) %}
-{%- for level in range(size, 1, -1) %}
-{%- if (size % level) == 0 %}
-{% if (level != size) %}
-{{gm.ident(1)}}for (int i=0; i<{{level}}; i++) {
-{{gm.ident(2)}}#pragma HLS unroll
-{{gm.ident(2)}}inter{{level}}[0][i]=inter{{level*2}}[0][i*2]+inter{{level*2}}[0][i*2+1];
-{{gm.ident(2)}}inter{{level}}[1][i]=inter{{level*2}}[1][i*2]+inter{{level*2}}[1][i*2+1];
-{{gm.ident(1)}}}
-{%- endif %}
-{%- endif %}
-{%- endfor %}
-{%- endmacro %}
-
-{# Description: Handle preprocess -- NOT USED #}
-{# ------ #}
-{# Params: #}
-{%- macro handle_preprocess(object, name) -%}
- {%- if name in object.keys() %}
- {%- for key, value in object.iteritems() %}
- {%- if key == name %}
- {%- if value > 0 %}
- {%- endif %}
- {%- endif %}
- {%- endfor %}
- {%- endif %}
-{%- endmacro %}
-
-{# Description: declare the layer functions--NOT USED #}
-{# ------ #}
-{# Params: #}
-{%- macro fdecl(object) -%}
-{%- endmacro %}
-
-{# Description: declare global variabels -- NOT USED #}
-{# ------ #}
-{# Params: #}
-{%- macro var_decl(object) -%}
-{%- endmacro %}
-
-{# Description: call validation function #}
-{# ------ #}
-{# Params: #}
-{# object -- layer object #}
-{% macro fvalid_call(object) %}
-{{gm.ident(1)}}{{object.name}}_test_fc();
-{% if object.activation == 'ReLU' %}
-{{gm.ident(1)}}{{object.name}}_test_relu();
-{% endif %}
-{% endmacro %}
-
-{# Description: call layer and activation functions #}
-{# ------ #}
-{# Params: #}
-{# object -- layer object #}
-{# flags -- flags for debugging, verifcation mode, etc. #}
-{%- macro fcall(object, prev, flags) -%}
-{%- set ip_shift = object.preprocess.ip_shift-%}
-{%- set int_shift = object.preprocess.int_shift-%}
-{%- set w_shift = object.weights.W_shift-%}
-{%- set op_shift = object.preprocess.op_shift -%}
-{%- set b_shift = object.weights.b_shift-%}
-{%- set op_loop_factor = object.op_loop_factor -%}
-{%- set ip_loop_factor = object.ip_loop_factor -%}
-{%- set right_shift = (ip_shift+w_shift-int_shift) -%}
-{%- set right_shift_relu = int_shift - op_shift -%}
-{%- set left_shift_relu = int_shift - b_shift-%}
-{# read weights from file and put into memory if in verification mode #}
-{%- if flags.verify -%}
-//previous layer is {{prev.name}}
-{{gm.ident(1)}}{{aux.fcall_readW_fc(object, prev)}}
-{% endif %}
-{{gm.ident(1)}}{{object.name}}_layer({{(object.dshape[0]/ip_loop_factor)|round(0,'ceil')|int}}, {{(object.wshape[1]/op_loop_factor)|round(0,'ceil')|int}}, {{right_shift}}, w,
-{# placeholder, in case we need to be able to switch between offchip memory and onchip memory #}
-{%- if object.interface.in == 'dram' %}
- a,
-{%- elif object.interface.in == 'onchip' %}
- a_onchip,
-{%- endif %}
-{%- if object.interface.int == 'dram' %}
- c
-{%- elif object.interface.int == 'onchip' %}
- c_onchip
-{%- endif %}
-);
-{# if activation function is required instantiate accordingly #}
-{% if object.activation == 'ReLU' %}
-{# read bias as needed #}
-{%- if flags.verify -%}
-{{gm.ident(1)}}{{aux.fcall_readb(object)}}
-{% endif %}
-{{gm.ident(1)}}{{object.name}}_bias_relu({{(object.wshape[1]/op_loop_factor)|round(0,'ceil')|int}}, {{left_shift_relu}}, {{right_shift_relu}}, w,
-{%- if object.interface.in == 'dram' %}
- c,
-{%- elif object.interface.in == 'onchip' %}
- c_onchip,
-{%- endif %}
-{%- if object.interface.int == 'dram' %}
- a
-{%- elif object.interface.int == 'onchip' %}
- a_onchip
-{%- endif %}
-);
-{% endif %}
-{%- endmacro %}
-
-{# Description: implement the memory read & unpack function #}
-{# ------ #}
-{# Params: #}
-{# layer -- name of the layer #}
-{# name -- name of the buffer to read: output/itrm,etc. #}
-{# mem_bw -- bit width of a single memory line #}
-{# bw -- bit width of the data we want to read #}
-{# loop_factor -- loop_factor, also the pack size #}
-{# TODO: share instances of read function between layers #}
-{% macro readFunc(layer, name, mem_bw, bw, loop_factor)%}
-void {{layer}}_read_{{name}}(
- {{gm.Xfui(mem_bw)}} *mem,
- {{gm.Xfixed_arr(bw-1, 0, 'inputToLayer', loop_factor)}},
- int address) {
-{% if mem_bw >= (loop_factor*bw) %}
-{% call gm.forloop(1, 'i', loop_factor) %}
-{%- if bw == 8-%}
-{{gm.ident(2)}}inputToLayer[i]={{gm.Xfixed(bw-1,0)}}::createRaw(mem[address].getChar(i));
-{%- elif bw == 16 -%}
-{{gm.ident(2)}}inputToLayer[i]={{gm.Xfixed(bw-1,0)}}::createRaw(mem[address].getShort(i));
-{%- endif -%}
-{% endcall %}
-{%- else -%}
-{% set mem_num = (mem_bw/bw)|int %}
-{% call gm.forloop_s(1, 'i', loop_factor, mem_num) %}
-{% call gm.forloop(2, 'j', mem_num) %}
-{%- if bw == 8-%}
-{{gm.ident(3)}}inputToLayer[i+j]={{gm.Xfixed(bw-1,0)}}::createRaw(mem[address].getChar(i));
-{%- elif bw == 16 -%}
-{{gm.ident(3)}}inputToLayer[i+j]={{gm.Xfixed(bw-1,0)}}::createRaw(mem[address].getShort(i));
-{%- endif -%}
-{%- endcall -%}
-{%- endcall -%}
-{%- endif -%}
-}
-{% endmacro %}
-
-{# Description: implement the memory write & pack function #}
-{# ------ #}
-{# Params: #}
-{# layer -- name of the layer #}
-{# name -- name of the buffer to write: output/itrm,etc.#}
-{# mem_bw -- bit width of a single memory line #}
-{# bw -- bit width of the data we want to read #}
-{# loop_factor -- loop_factor, also the pack size #}
-{# TODO: share instances of read function between layers #}
-{% macro writeFunc(layer, name, mem_bw, bw, loop_factor)%}
-void {{layer}}_write_{{name}}(
- {{gm.Xfui(mem_bw)}} *mem,
- {{gm.Xfixed_arr(bw-1, 0, 'outputFromLayer', loop_factor)}},
- int address) {
-{% if mem_bw >= (loop_factor*bw) %}
-{% call gm.forloop(1, 'i', loop_factor) %}
-{% if bw == 16 %}
-{{gm.ident(2)}}mem[address].setValue(i, outputFromLayer[i].getRaw());
-{% elif bw == 8 %}
-{{gm.ident(2)}}mem[address].setValue(i, (char)outputFromLayer[i].getRaw());
-{% endif %}
-{% endcall %}
-{%- else -%}
-{% set mem_num = (mem_bw/bw)|int %}
-{% call gm.forloop_s(1, 'i', loop_factor, mem_num) %}
-{% call gm.forloop(2, 'j', mem_num) %}
-{% if bw == 16 %}
-{{gm.ident(3)}}mem[address].setValue(j, outputFromLayer[i+j].getRaw());
-{% elif bw == 8 %}
-{{gm.ident(3)}}mem[address].setValue(j, (char)outputFromLayer[i+j].getRaw());
-{% endif %}
-{%- endcall -%}
-{%- endcall -%}
-{%- endif -%}
-}
-{% endmacro %}
-
-{# Description: implement the layer functions #}
-{# #}
-{# ------ #}
-{# Params: #}
-{# object: layer object that contains all the info #}
-{%- macro fimpl(object) -%}
-{%- set ip_bw = object.preprocess.ip_bit_width-%}
-{%- set ip_shift = object.preprocess.ip_shift-%}
-{%- set int_bw = object.preprocess.int_bit_width-%}
-{%- set int_shift = object.preprocess.int_shift-%}
-{%- set op_bw = object.preprocess.op_bit_width-%}
-{%- set op_shift = object.preprocess.op_shift -%}
-{# parameters for clustering to define if clustering is enabled #}
-{% if object.preprocess.cluster == true %}
-{%- set index_width = object.preprocess.index_width -%}
-{%- set num_clusters = object.preprocess.num_clusters -%}
-{% endif %}
-{%- set w_bw = object.weights.W_width-%}
-{%- set w_shift = object.weights.W_shift-%}
-{%- set b_bw = object.weights.b_width-%}
-{%- set b_shift = object.weights.b_shift-%}
-{%- set ip_loop_factor = object.ip_loop_factor -%}
-{%- set op_loop_factor = object.op_loop_factor -%}
-{%- set mem_bw = object.interface.memory_bit_width -%}
-{# internal bw is the "safe" bit width for the result of multiplication #}
-{%- set internal_bw = w_bw+ip_bw -%}
-{%- set iters_weight = (w_bw*ip_loop_factor*2/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_bias = (ip_bw*ip_loop_factor/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_input = (ip_bw*ip_loop_factor/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_output = (op_bw*op_loop_factor/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_itrm = (int_bw*op_loop_factor/mem_bw)|round(0,'ceil')|int -%}
-
-{# Implement the memory read & unpack function #}
-{# see the overview doc for more detail #}
-/*
-Name: {{object.name}}_read_weight
-Description: read from memory and write to the two weight buffers.
- each of the buffers holds one 'pack' of 'loop_factor'
- numbers of data. Notice that we read two packs in at
- the same time because the compengine will consume
- two weight packs in one iteration. When clustering is
- enabled, the input bit-width will be replaced by the
- bit-width of the indices.
-Params:
- mem -- the memory interface to read from
- weight1 -- weight pack 1, to be passed to the compengine
- weight2 -- weight pack 2, to be passed to the compengine
-*/
-{% if object.preprocess.cluster == true%}
-void {{object.name}}_read_weight(
- {{gm.Xfui(mem_bw)}} *mem,
- {{gm.Xfixed_arr(index_width-1, 0, 'weight1', ip_loop_factor)}},
- {{gm.Xfixed_arr(index_width-1, 0, 'weight2', ip_loop_factor)}},
- int address) {
-{%- if mem_bw >= (2*ip_loop_factor*index_width) -%}
-{% call gm.forloop(1, 'i', ip_loop_factor) %}
-{%- if index_width == 8-%}
-{{gm.ident(2)}}weight1[i]={{gm.Xfixed(index_width-1,0)}}::createRaw(mem[address].getChar(i));
-{{gm.ident(2)}}weight2[i]={{gm.Xfixed(index_width-1,0)}}::createRaw(mem[address].getChar(i+{{ip_loop_factor}}));
-{%- elif index_width == 16 -%}
-{{gm.ident(2)}}weight1[i]={{gm.Xfixed(index_width-1,0)}}::createRaw(mem[address].getChar(i));
-{{gm.ident(2)}}weight2[i]={{gm.Xfixed(index_width-1,0)}}::createRaw(mem[address].getChar(i+{{ip_loop_factor}}));
-{%- endif -%}
-{% endcall %}
-{%- else -%}
-{%- set mem_num = (mem_bw/2/index_width)|int -%}
-{%- call gm.forloop_s(1, 'i', ip_loop_factor, mem_num) -%}
-{%- call gm.forloop(2, 'j', mem_num) -%}
-{%- if index_width == 8-%}
-{{gm.ident(3)}}weight1[i+j]={{gm.Xfixed(index_width-1,0)}}::createRaw(mem[address].getChar(i));
-{{gm.ident(3)}}weight2[i+j]={{gm.Xfixed(index_width-1,0)}}::createRaw(mem[address].getChar(i+{{mem_num}}));
-{%- elif index_width == 16 -%}
-{{gm.ident(3)}}weight1[i+j]={{gm.Xfixed(index_width-1,0)}}::createRaw(mem[address].getChar(i));
-{{gm.ident(3)}}weight2[i+j]={{gm.Xfixed(index_width-1,0)}}::createRaw(mem[address].getChar(i+{{mem_num}}));
-{%- endif -%}
-{%- endcall -%}
-{%- endcall -%}
-{%- endif -%}
-}
-{% else %}
-void {{object.name}}_read_weight(
- {{gm.Xfui(mem_bw)}} *mem,
- {{gm.Xfixed_arr(w_bw-1, 0, 'weight1', ip_loop_factor)}},
- {{gm.Xfixed_arr(w_bw-1, 0, 'weight2', ip_loop_factor)}},
- int address) {
-{%- if mem_bw >= (2*ip_loop_factor*w_bw) -%}
-{% call gm.forloop(1, 'i', ip_loop_factor) %}
-{%- if w_bw == 8-%}
-{{gm.ident(2)}}weight1[i]={{gm.Xfixed(w_bw-1,0)}}::createRaw(mem[address].getChar(i));
-{{gm.ident(2)}}weight2[i]={{gm.Xfixed(w_bw-1,0)}}::createRaw(mem[address].getChar(i+{{ip_loop_factor}}));
-{%- elif w_bw == 16 -%}
-{{gm.ident(2)}}weight1[i]={{gm.Xfixed(w_bw-1,0)}}::createRaw(mem[address].getChar(i));
-{{gm.ident(2)}}weight2[i]={{gm.Xfixed(w_bw-1,0)}}::createRaw(mem[address].getChar(i+{{ip_loop_factor}}));
-{%- endif -%}
-{% endcall %}
-{%- else -%}
-{%- set mem_num = (mem_bw/2/w_bw)|int -%}
-{%- call gm.forloop_s(1, 'i', ip_loop_factor, mem_num) -%}
-{%- call gm.forloop(2, 'j', mem_num) -%}
-{%- if w_bw == 8-%}
-{{gm.ident(3)}}weight1[i+j]={{gm.Xfixed(w_bw-1,0)}}::createRaw(mem[address].getChar(i));
-{{gm.ident(3)}}weight2[i+j]={{gm.Xfixed(w_bw-1,0)}}::createRaw(mem[address].getChar(i+{{mem_num}}));
-{%- elif w_bw == 16 -%}
-{{gm.ident(3)}}weight1[i+j]={{gm.Xfixed(w_bw-1,0)}}::createRaw(mem[address].getChar(i));
-{{gm.ident(3)}}weight2[i+j]={{gm.Xfixed(w_bw-1,0)}}::createRaw(mem[address].getChar(i+{{mem_num}}));
-{%- endif -%}
-{%- endcall -%}
-{%- endcall -%}
-{%- endif -%}
-}
-{% endif %}
-
-{{readFunc(object.name, "input", mem_bw, ip_bw, ip_loop_factor)}}
-{{readFunc(object.name,"itrm", mem_bw, int_bw, op_loop_factor)}}
-{{readFunc(object.name,"bias", mem_bw, b_bw, op_loop_factor)}}
-{{writeFunc(object.name,"itrm", mem_bw, int_bw, op_loop_factor)}}
-{{writeFunc(object.name,"output", mem_bw, op_bw, op_loop_factor)}}
-
-{% if object.preprocess.cluster == true %}
-/*
-Name: {{object.name}}_weight_table
-Description: store the lookup table for clustering and performs the lookup
-Params: the parameters
-Return: the actual weight get after lookup
-*/
-{{gm.Xfixed(w_bw-1, 0)}} weightTable(int index)
-{
-{{gm.ident(1)}}{{gm.Xfixed_arr(w_bw-1, 0, 'table', num_clusters)}};
-{%- for weight in object.cluster_values -%}
-{{gm.ident(1)}}table[{{loop.index-1}}]={{weight}};
-{% endfor %}
-{{gm.ident(1)}}return table[index];
-}
-
-/*
-Name: {{object.name}}_decode_weight
-Description: Decode a pack of indices and put into an array of weights, it will call the *_weight_table function
-Params:
- weightArray -- an array of weights that can be dierectly
- indexPack -- an array of indicies that needs to be decoded with the lookup table
-*/
-void {{object.name}}_decode_weight({{gm.Xfixed_arr(w_bw-1, 0, 'weightArray', ip_loop_factor)}}, {{gm.Xfixed_arr(index_width-1, 0, 'indices', ip_loop_factor)}})
-{
-{% call gm.forloop(1, 'i', ip_loop_factor) %}
-{{gm.ident(2)}}int indexTmp;
-{{gm.ident(2)}}indexTmp = indices[i].getRaw();
-{{gm.ident(2)}}weightArray[i]=weightTable(indexTmp);
-{%- endcall -%}
-}
-{% endif %}
-
-/*
-Name: {{object.name}}_saturate_add
-Description: this function performs add with a saturation,
- it checks the result of addition and clamp the
- output to MAX/MIN value of the datatype
-Params: a -- input value a
- b -- input value b
-Return: c -- output value c after saturation
-*/
-{{gm.Xfixed(int_bw-1,0)}}{{object.name}}_saturate_add(
-{{gm.Xfixed_vdecl(int_bw-1, 0, 'a')}},
-{{gm.Xfixed_vdecl(int_bw-1, 0, 'b')}})
-{
-{{ident(1)}}float OVERFLOWMAX = {{2.0**(int_bw - 1) - 1.0}};
-{{ident(1)}}float OVERFLOWMIN = -{{2.0**(int_bw - 1)}};
-
-{{ident(1)}}{{gm.Xfixed_vdecl(int_bw-1, 0, 'c')}};
-{{ident(1)}}c = a + b;
-{{ident(1)}}if((a > 0)&&(b > 0)&&(c < 0)){
-{{ident(2)}}c = OVERFLOWMAX;
-{{ident(1)}}} else if ((a < 0)&&(b < 0)&&(c > 0)) {
-{{ident(2)}}c = OVERFLOWMIN;
-{{ident(1)}}}
-{{ident(1)}}return c;
-}
-
-{# determine whether multiplication can be packed #}
-/*
-Name: {{object.name}}_pack_dsp
-Description: Core arithmetic operation, do two multiplications and store in rst1 & rst2.
- NOTE: for now (8/16/2018) this is just a simple version. In the future we
- may utilize the function of DSP slices of the FPGA (depending on what FPGA
- we want ot use) to pack two multiplications in one operation.
-Params:
- rst1 -- result 1
- rst2 -- result 2
- a -- multiplicand - from weight
- b -- multiplicand - from weight
- c -- multiplier - from input
-*/
-void {{object.name}}_pack_dsp (
- {{gm.Xfixed_vdecl(int_bw-1, 0, '&rst1')}},
- {{gm.Xfixed_vdecl(int_bw-1, 0, '&rst2')}},
- {{gm.Xfixed_vdecl(w_bw-1, 0, 'a')}},
- {{gm.Xfixed_vdecl(w_bw-1, 0, 'b')}},
- {{gm.Xfixed_vdecl(ip_bw-1, 0, 'c')}}
-) {
- rst1 = a * c;
- rst2 = b * c;
-}
-
-/*
-Name: {{object.name}}_compengine
-Description: Computational engine that multiply two packs of weights with one packs of input,
- compute the sum of the result of multiplications and generate two numbers as resuts.
-Params:
- out1 -- result 1
- out2 -- result 2
- in -- array of numbers from input
- weight1 -- array of weights 1
- weight2 -- array of weights 2
-*/
-void {{object.name}}_compengine_row(
- {{gm.Xfixed_vdecl(int_bw-1, 0,'&out1')}},
- {{gm.Xfixed_vdecl(int_bw-1, 0,'&out2')}},
- {{gm.Xfixed_arr(ip_bw-1, 0, 'in', ip_loop_factor)}},
-{% if object.preprocess.cluster == true %}
- {{gm.Xfixed_arr(index_width-1, 0, 'indices1', ip_loop_factor)}},
- {{gm.Xfixed_arr(index_width-1, 0, 'indices2', ip_loop_factor)}}
-{%- else %}
- {{gm.Xfixed_arr(w_bw-1, 0, 'weight1', ip_loop_factor)}},
- {{gm.Xfixed_arr(w_bw-1, 0, 'weight2', ip_loop_factor)}}
-{%- endif %}
-) {
-
-{% if object.preprocess.cluster == true %}
- {{gm.Xfixed_arr(w_bw-1, 0, 'weight1', ip_loop_factor)}};
- {{gm.Xfixed_arr(w_bw-1, 0, 'weight2', ip_loop_factor)}};
-{% endif %}
-{{gen_red_tree_vdecl(ip_loop_factor, gm.Xfixed(int_bw-1,0))}}
-{# if cluster is enabled #}
-{%- if object.preprocess.cluster == true -%}
-{{gm.ident(1)}}//Decode clustered input
-{{gm.ident(1)}}{{object.name}}_decode_weight(weight1, indices1);
-{{gm.ident(1)}}{{object.name}}_decode_weight(weight2, indices2);
-{% endif %}
-{%- call gm.forloop(1, 'i', ip_loop_factor) %}
-{{gm.ident(2)}}#pragma HLS unroll
-{{gm.ident(2)}}{{object.name}}_pack_dsp(inter{{ip_loop_factor}}[0][i],inter{{ip_loop_factor}}[1][i],weight1[i],weight2[i],in[i]);
-{%- endcall %}
-{{gen_red_tree_impl(ip_loop_factor, gm.Xfixed(int_bw-1,0))}}
- out1=inter2[0][0]+inter2[0][1];
- out2=inter2[1][0]+inter2[1][1];
-}
-
-/*
-Name: {{object.name}}_out_converge
-Description: Accumulation operation that add the partial result of current iteration to
- the final result.
-Params: outPack -- final result of the convolution
- outTemp -- output from the current iteration (for 1 pixel in the filter)
-*/
-void {{object.name}}_out_converge({{gm.Xfixed_arr(int_bw-1, 0, 'outPack', op_loop_factor)}}, {{gm.Xfixed_arr(int_bw-1, 0,'outTemp',op_loop_factor)}}, int right_shift) {
-{% call gm.forloop(1,'i',op_loop_factor) %}
-{{gm.ident(2)}}outPack[i]={{object.name}}_saturate_add(outPack[i], (outTemp[i] >> right_shift));
-{%- endcall -%}
-}
-
-/*
-Name: {{object.name}}_layer
-Description: top level function of the layer. Since fully connected layer
- can be seen as a special case of fully connected layer. The
- structure of them are very similar. The difference is that in
- FC layer, we removed the loop for height and weight, because
- for FC they are both 1.
-*/
-void {{object.name}}_layer(
- int CII,//image channels
- int COO, //output channels
- int right_shift,
- {{gm.Xfui_vdecl(mem_bw, '*weight')}},
- {%- if object.interface.in == 'dram' %}
- {{gm.Xfui_vdecl(mem_bw, '*in')}},
- {%- elif object.interface.in == 'onchip' %}
- {{gm.Xfui_vdecl_arr(onchip_bw, 'in', 50176)}},
- {%- endif %}
- {%- if object.interface.int == 'dram' %}
- {{gm.Xfui_vdecl(mem_bw, '*out')}}
- {%- elif object.interface.int == 'onchip' %}
- {{gm.Xfui_vdecl_arr(onchip_bw, 'out', 50176)}}
- {%- endif %}
-){
-{{gm.ident(1)}}//Clear initial values of output buffers
-{% if (object.interface.int == 'dram'): %}
-{% call gm.forloop(1, 'i', 'COO*%d'|format(iters_itrm)) %}
-{{gm.ident(2)}}{{gm.Xfui_vdecl(mem_bw, 'temp')}}((unsigned int) 0);
-{{gm.ident(2)}}out[i] = temp;
-{% endcall %}
-{% elif (object.interface.int == 'onchip'): %}
-{% call gm.forloop(1, 'i', 'COO*%d'|format(iters_itrm)) %}
-{{gm.ident(2)}}{{gm.Xfui_vdecl(onchip_bw, 'temp')}}((unsigned int) 0);
-{{gm.ident(2)}}out[i] = temp;
-{% endcall %}
-{% endif %}
-{{gm.ident(1)}}int weight_idx=0;
-{{gm.ident(1)}}//First set of loops: Iterate over filter
-{% call gm.forloop(1, 'cii', 'CII') %}
-{% call gm.forloop(2, 'coo', 'COO') %}
-{{gm.ident(3)}}{{gm.Xfixed_arr(ip_bw-1, 0,'inTemp', ip_loop_factor)}};
-{{gm.ident(3)}}{{gm.Xfixed_arr(int_bw-1, 0,'outTemp', op_loop_factor)}};
-{% if object.interface.in == 'dram' %}
-{{gm.ident(3)}}int address=cii*{{iters_input}};
-{{gm.ident(3)}}{{object.name}}_read_input(in, inTemp, address);
-{% elif object.interface.in == 'onchip' %}
-{{gm.ident(3)}}int address=cii*{{iters_input}};
-{{gm.ident(3)}}{{object.name}}_read_input(in, inTemp, address);
-{% endif %}
-{#if the output length is shorter than the op_loop_factor#}
-{% if (object.wshape[1] >= op_loop_factor)%} {# op_loop_factor #}
-{% call gm.forloop_s(3, 'i', op_loop_factor, 2) %}
-{% if object.preprocess.cluster == true %}
-{{gm.ident(4)}}{{gm.Xfixed_arr(index_width-1, 0, 'weight1', ip_loop_factor)}};
-{{gm.ident(4)}}{{gm.Xfixed_arr(index_width-1, 0, 'weight2', ip_loop_factor)}};
-{% else %}
-{{gm.ident(4)}}{{gm.Xfixed_arr(w_bw-1, 0, 'weight1', ip_loop_factor)}};
-{{gm.ident(4)}}{{gm.Xfixed_arr(w_bw-1, 0, 'weight2', ip_loop_factor)}};
-{% endif%}
-{{gm.ident(4)}}int address = weight_idx+i/2*{{iters_weight}};
-{{gm.ident(4)}}{{object.name}}_read_weight(weight, weight1, weight2, address);
-{{gm.ident(4)}}{{object.name}}_compengine_row(outTemp[i], outTemp[i+1], inTemp, weight1, weight2);
-{%- endcall -%}
-{{gm.ident(3)}}{{gm.Xfixed_arr(int_bw-1, 0, 'outPackTemp', op_loop_factor)}};
-{% if object.interface.int == 'dram' %}
-{{gm.ident(3)}}address=coo*{{iters_itrm}};
-{{gm.ident(3)}}{{object.name}}_read_itrm(out, outPackTemp, address);
-{{gm.ident(3)}}{{object.name}}_out_converge(outPackTemp, outTemp, right_shift);
-{{gm.ident(3)}}{{object.name}}_write_itrm(out, outPackTemp, address);
-{{gm.ident(3)}}weight_idx += ({{op_loop_factor}}/2)*{{iters_weight}};
-{% elif object.interface.int == 'onchip' %}
-{{gm.ident(3)}}address=coo*{{iters_itrm}};
-{{gm.ident(3)}}{{object.name}}_read_itrm(out, outPackTemp, address);
-{{gm.ident(3)}}{{object.name}}_out_converge(outPackTemp, outTemp, right_shift);
-{{gm.ident(3)}}{{object.name}}_write_itrm(out, outPackTemp, address);
-{{gm.ident(3)}}weight_idx += ({{op_loop_factor}}/2)*{{iters_weight}};
-{% endif %}
-{# if the channels of output is smaller than loop factor #}
-{# it need some special attention #}
-{% else %}
-{% call gm.forloop_s(4, 'i', object.wshape[1], 2) %}
-{% if object.preprocess.cluster == true %}
-{{gm.ident(8)}}{{gm.Xfixed_arr(index_width-1, 0, 'weight1', ip_loop_factor)}};
-{{gm.ident(8)}}{{gm.Xfixed_arr(index_width-1, 0, 'weight2', ip_loop_factor)}};
-{% else %}
-{{gm.ident(8)}}{{gm.Xfixed_arr(w_bw-1, 0, 'weight1', ip_loop_factor)}};
-{{gm.ident(8)}}{{gm.Xfixed_arr(w_bw-1, 0, 'weight2', ip_loop_factor)}};
-{% endif%}
-{{gm.ident(8)}}int address = weight_idx+i/2*{{iters_weight}};
-{{gm.ident(8)}}{{object.name}}_read_weight(weight, weight1, weight2, address);
-{{gm.ident(8)}}{{object.name}}_compengine_row(outTemp[i], outTemp[i+1], inTemp, weight1, weight2);
-{% endcall %}
-{% if object.interface.int == 'dram' %}
-{{gm.ident(7)}}{{gm.Xfixed_arr(int_bw-1, 0, 'outPackTemp', op_loop_factor)}};
-{{gm.ident(3)}}address=coo*{{iters_itrm}};
-{{gm.ident(3)}}{{object.name}}_read_itrm(out, outPackTemp, address);
-{{gm.ident(3)}}{{object.name}}_out_converge(outPackTemp, outTemp, right_shift);
-{{gm.ident(3)}}{{object.name}}_write_itrm(out, outPackTemp, address);
-{{gm.ident(3)}}weight_idx += ({{object.wshape[1]}}/2)*{{iters_weight}};
-{% elif object.interface.int == 'onchip' %}
-{{gm.ident(7)}}{{gm.Xfixed_arr(int_bw-1, 0, 'outPackTemp', op_loop_factor)}};
-{{gm.ident(3)}}address=coo*{{iters_itrm}};
-{{gm.ident(3)}}{{object.name}}_read_itrm(out, outPackTemp, address);
-{{gm.ident(3)}}{{object.name}}_out_converge(outPackTemp, outTemp, right_shift);
-{{gm.ident(3)}}{{object.name}}_write_itrm(out, outPackTemp, address);
-{{gm.ident(3)}}weight_idx += ({{object.wshape[1]}}/2)*{{iters_weight}};
-{%- endif -%}
-{%- endif -%}
-{%- endcall -%}
-{%- endcall -%}
-}
-
-{% if object.activation == 'ReLU' %}
-/*
-Name: {{object.name}}_saturate_relu
-Description: performs saturation when converting from int_bw to op_bw
-Params:
- in -- number to be converted to output bit-width
-Return:
- temp -- result after saturation in output bit-width
-*/
-{{gm.Xfixed(op_bw-1,0)}} {{object.name}}_saturate_relu(
-{{gm.ident(1)}}{{gm.Xfixed_vdecl(int_bw-1, 0, 'in')}})
-{
-{{gm.ident(1)}}float OVERFLOWMAX = {{2.0**(op_bw - 1) - 1.0}};
-{{gm.ident(1)}}float OVERFLOWMIN = -{{2.0**(op_bw - 1)}};
-
-{{gm.ident(1)}}{{gm.Xfixed_vdecl(op_bw-1, 0, 'temp')}};
-{{gm.ident(1)}}if(in >= OVERFLOWMAX){
-{{gm.ident(2)}}temp = OVERFLOWMAX;
-{{gm.ident(1)}}} else if (in <= OVERFLOWMIN) {
-{{gm.ident(2)}}temp = OVERFLOWMIN;
-{{gm.ident(1)}}} else {
-{{gm.ident(2)}}temp=in;
-{{gm.ident(1)}}}
-{{gm.ident(1)}}return temp;
-}
-
-/*
-Name: {{object.name}}_bias_relu
-Description: computational kernel of bias relu layer,
- also performs saturation when adding bias
- to the pre-activation result as well as
- converting it to outputs with smaller
- bit-width.
-Params:
- in -- array of intermediate/pre-activation results
- bias -- array of biases
- out -- output of the bias & activation layer
-*/
-void {{object.name}}_bias_relu(
- {{gm.Xfixed_arr(int_bw-1, 0,'in',op_loop_factor)}},
- {{gm.Xfixed_arr(b_bw-1, 0,'bias',op_loop_factor)}},
- {{gm.Xfixed_arr(op_bw-1, 0,'out',op_loop_factor)}},
- int left_shift,
- int right_shift
-){
-{% call gm.forloop(1, 'j', op_loop_factor) %}
-{{gm.ident(2)}}{{gm.Xfixed_vdecl(int_bw-1, 0,'temp')}};
-{{gm.ident(2)}}{{gm.Xfixed_vdecl(int_bw-1, 0,'tempBias')}};
-{{gm.ident(2)}}tempBias = bias[j];
-{{gm.ident(2)}}temp={{object.name}}_saturate_add(in[j], (tempBias << left_shift));
-{{gm.ident(2)}}temp = temp >> right_shift;
-{{gm.ident(2)}}if(temp > 0)
-{{gm.ident(3)}}out[j] = {{object.name}}_saturate_relu(temp);
-{{gm.ident(2)}}else
-{{gm.ident(3)}}out[j] = 0.0;
-{%- endcall -%}
-}
-
-/*
-Name: {{object.name}}_bias_relu
-Description: top level function for the ReLU activation layer. This layer also
- performs "adding bias" and padding for the next layer.
-Params: see below
-*/
-void {{object.name}}_bias_relu(
- int COO,
- int left_shift,
- int right_shift,
- {{gm.Xfui_vdecl(mem_bw, '*weight')}},
- {%- if object.interface.in == 'dram' %}
- {{gm.Xfui_vdecl(mem_bw, '*in')}},
- {%- elif object.interface.in == 'onchip' %}
- {{gm.Xfui_vdecl_arr(onchip_bw, 'in', 50176)}},
- {%- endif %}
- {%- if object.interface.int == 'dram' %}
- {{gm.Xfui_vdecl(mem_bw, '*out')}}
- {%- elif object.interface.int == 'onchip' %}
- {{gm.Xfui_vdecl_arr(onchip_bw, 'out', 50176)}}
- {%- endif %}
-) {
-{{gm.ident(1)}}{{gm.Xfixed_arr(int_bw-1, 0, 'inTemp', op_loop_factor)}};
-{{gm.ident(1)}}{{gm.Xfixed_arr(op_bw-1, 0, 'outTemp', op_loop_factor)}};
-{{gm.ident(1)}}{{gm.Xfixed_arr(b_bw-1, 0, 'bias', op_loop_factor)}};
-{{gm.ident(1)}}int weight_idx = 0;
-{{gm.ident(1)}}int address = 0;
-{% call gm.forloop(1, 'co', 'COO') %}
-{{gm.ident(2)}}{{object.name}}_read_bias(weight, bias, co*{{iters_bias}});
-{{gm.ident(2)}}{{object.name}}_read_itrm(in, inTemp, co*{{iters_itrm}});
-{{gm.ident(2)}}{{object.name}}_bias_relu(inTemp, bias, outTemp,left_shift, right_shift);
-{{gm.ident(2)}}{{object.name}}_write_output(out, outTemp, co*{{iters_output}});
-{% endcall %}
-}
-{% endif %}
-{%- endmacro %}
-
-{% macro fvalid(object, prev, flags) %}
-{%- set IC = object.wshape[0] -%}
-{%- set OC = object.wshape[1] -%}
-{%- set ip_loop_factor = object.ip_loop_factor -%}
-{%- set op_loop_factor = object.op_loop_factor -%}
-{# ---------------------------- #}
-{# the following code determins the correct size of output of previous layer#}
-{% if prev.layer_type == 'fc' %}
-{%- set H = 1 -%}
-{%- set W = 1 -%}
-{%- set C = prev.wshape[1] -%}
-{% elif prev.layer_type == 'max_pool' %}
-{%- set stride = prev.weights.stride[1] -%}
-{%- set ksize = prev.weights.ksize[1] -%}
-{%- set padsize = prev.weights.padsize -%}
-{%- set C = prev.dshape[2] -%}
-{% if prev.mode == 'SAME' %}
-{%- set H = (prev.dshape[0]/stride)|round(0,'ceil')|int -%}
-{%- set W = (prev.dshape[1]/stride)|round(0,'ceil')|int -%}
-{% else %}
-{%- set H = ((prev.dshape[0]-ksize+1)/stride)|round(0,'ceil')|int -%}
-{%- set W = ((prev.dshape[1]-ksize+1)/stride)|round(0,'ceil')|int -%}
-{% endif %}
-{% endif %}
-{#----------------------------#}
-{%- set ip_bw = object.preprocess.ip_bit_width-%}
-{%- set int_bw = object.preprocess.int_bit_width-%}
-{%- set op_bw = object.preprocess.op_bit_width-%}
-{%- set w_bw = object.weights.W_width-%}
-{%- set b_bw = object.weights.b_width-%}
-{%- set mem_bw = object.interface.memory_bit_width -%}
-{# internal bw is the "safe" bit width for the result of multiplication #}
-{%- set iters_weight = (w_bw*ip_loop_factor*2/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_bias = (ip_bw*ip_loop_factor/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_input = (ip_bw*ip_loop_factor/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_output = (op_bw*op_loop_factor/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_itrm = (int_bw*op_loop_factor/mem_bw)|round(0,'ceil')|int -%}
-
-void {{object.name}}_test_fc() {
- //test framework
- srand(17);
- bool flag = false;
-//Allocate buffers for test data
-{{gm.ident(1)}}int* input_buf = new int[{{IC}}];
-{{gm.ident(1)}}int* weight_buf = new int[{{IC}}*{{OC}}];
-{{gm.ident(1)}}int* output_buf = new int[{{OC}}];
-//Generate random input
-{% call gm.forloop(1, 'c', IC) %}
-{{gm.ident(2)}}input_buf[c] = rand()%2;
-{%- endcall -%}
-
-//transform HWC order into CHW order
-{% call gm.forloop(1, 'c', (C/ip_loop_factor)|round(0,'ceil')|int) %}
-{% call gm.forloop(2, 'h', H) %}
-{% call gm.forloop(3, 'w', W) %}
-{% call gm.forloop(4, 'ci', ip_loop_factor) %}
-{{gm.ident(5)}}((ap_int<512>*)a)[c*{{H}}*{{W}}+h*{{W}}+w].{{gm.pr_range_idx(ip_bw, 'ci')}} = input_buf[h*{{W}}*{{C}}+w*{{C}}+(c*{{ip_loop_factor}}+ci)];
-{%- endcall -%}
-{%- endcall -%}
-{%- endcall -%}
-{%- endcall -%}
-
-//generate random weights and pack
-{% call gm.forloop(1, 'i', OC) %}
-{% call gm.forloop(2, 'j', IC) %}
-{{gm.ident(2)}}weight_buf[i*{{IC}}+j] = rand()%2;
-{% endcall %}
-{% endcall %}
-pack_weights_fc<int>({{H}}, {{W}}, {{C}}, {{OC}}, weight_buf, (ap_int<512>*)w);
-
-//compute gold
-fc(output_buf, weight_buf, input_buf, {{IC}}, {{OC}});
-//compute real
-{{object.name}}_layer({{(IC/ip_loop_factor)|round(0,'ceil')|int}}, {{(OC/op_loop_factor)|round(0,'ceil')|int}}, 0, w, a, c);
-
-//compare
-{% call gm.forloop(1, 'i', (OC/op_loop_factor)|round(0,'ceil')|int) %}
-{% call gm.forloop(2, 'ii', [OC, op_loop_factor]|min) %}
-{{gm.ident(5)}}{{gm.fi_vdecl(mem_bw, 'temp_c')}} = ((ap_int<512>*)c)[i*{{iters_itrm}}];
-{{gm.ident(5)}}{{gm.fi_vdecl(int_bw, 'out_layer')}} = temp_c.{{gm.pr_range_idx(16,'ii')}};
-{{gm.ident(5)}}int out_golden = output_buf[(i*{{op_loop_factor}}+ii)];
-{{gm.ident(5)}}if (out_layer != out_golden) {
-{{gm.ident(5)}}cout << "[" << (i*{{op_loop_factor}}+ii) << "]";
-{{gm.ident(5)}}cout << "Diff: Layer - " << out_layer << "Golden - " << out_golden << endl;
-{{gm.ident(5)}}flag = true;
-{{gm.ident(5)}}}
-{%- endcall -%}
-{{gm.ident(2)}}//cout << endl;
-{%- endcall -%}
-
-{{gm.ident(1)}}delete [] input_buf;
-{{gm.ident(1)}}delete [] output_buf;
-{{gm.ident(1)}}delete [] weight_buf;
-{{gm.ident(1)}}assert(!flag);
-}
-
-{% if object.activation == 'ReLU' %}
-void {{object.name}}_test_relu() {
-//Dummy function
-//ReLU for FC is very simple so we omitted the validation function for now
-}
-{%- endif -%}
-{% endmacro %}
\ No newline at end of file
diff --git a/DRED_codegen/templates/global_macros.inc b/DRED_codegen/templates/global_macros.inc
deleted file mode 100644
index 9ce4e993c710b85f8d920e28983660e126f30bef..0000000000000000000000000000000000000000
--- a/DRED_codegen/templates/global_macros.inc
+++ /dev/null
@@ -1,131 +0,0 @@
-{# indentation #}
-{%- macro ident(lvl) %}
-{% for i in range(lvl) %} {% endfor %}
-{%- endmacro %}
-
-{# basic for loop, start from 0 #}
-{% macro forloop(lvl, idx, bound) %}
-{{ident(lvl)}}for (int {{idx}}=0; {{idx}} < {{bound}}; {{idx}}++) {
-{{ caller() }}
-{{ident(lvl)}}}
-{% endmacro %}
-
-{# for loop with non-uniform stride #}
-{% macro forloop_s(lvl, idx, bound, stride) %}
-{{ident(lvl)}}for (int {{idx}}=0; {{idx}} < {{bound}}; {{idx}}+={{stride}}) {
-{{ caller() }}
-{{ident(lvl)}}}
-{% endmacro %}
-
-{# for loop with configurable boundaries and stride #}
-{% macro forloop_i(lvl, init, idx, bound, stride) %}
-{{ident(lvl)}}for (int {{idx}}={{init}}; {{idx}} < {{bound}}; {{idx}}+={{stride}}) {
-{{ caller() }}
-{{ident(lvl)}}}
-{% endmacro %}
-
-{# select bit range from 0 to len #}
-{%- macro pr_range(len) -%}
-range({{len - 1}},0)
-{%- endmacro %}
-
-{# select bit range from starat to end #}
-{%- macro pr_range_bound(start,end) -%}
-range({{start}},{{end}})
-{%- endmacro %}
-
-{# select bit range from starat to end wiht index #}
-{%- macro pr_range_idx(len, id) -%}
-range({{len}}*{{id}}+{{len-1}},{{len}}*{{id}})
-{%- endmacro %}
-
-{# declare arbitrary length int #}
-{%- macro fixed_int(size) %}
-ap_int<{{size}}>
-{%- endmacro %}
-
-{# declare arbitrary length unsigned int #}
-{%- macro fixed_uint(size) %}
-ap_uint<{{size}}>
-{%- endmacro %}
-
-{# declare fixed point number with int_size number of integer bits #}
-{%- macro fixed(size, int_size) %}
-ap_fixed<{{size}},{{int_size}},AP_RND_CONV,AP_SAT>
-{%- endmacro %}
-
-{# declare an integer with parameterized name #}
-{%- macro fi_vdecl(size, name) %}
-{{fixed_int(size)}} {{name}}
-{%- endmacro %}
-
-{# declare an integer with parameterized name #}
-{%- macro fi_vdecl_arr(size, name, asize) %}
-{{fixed_int(size)}} {{name}}[{{asize}}]
-{%- endmacro %}
-
-{# declare an integer with parameterized name #}
-{%- macro fui_vdecl(size, name) %}
-{{fixed_uint(size)}} {{name}}
-{%- endmacro %}
-
-{# declare an array of unsigned integer with parameterized name #}
-{%- macro fui_vdecl_arr(size, name, asize) %}
-{{fixed_uint(size)}} {{name}}[{{asize}}]
-{%- endmacro %}
-
-{# declare an array of fixed point number #}
-{%- macro f_vdecl(size, int_size, name) %}
-{{fixed(size, int_size)}} {{name}}
-{%- endmacro %}
-
-{# declare an array of fixed point number with parameterized name#}
-{%- macro f_vdecl_arr(size, int_size, name, asize) %}
-{{fixed(size, int_size)}} {{name}}[{{asize}}]
-{%- endmacro %}
-
-{# declare an 2d array of fixed point number with parameterized name#}
-{%- macro f_vdecl_2darr(size, int_size, name, a1size, a2size) %}
-{{fixed(size, int_size)}} {{name}}[{{a1size}}][{{a2size}}]
-{%- endmacro %}
-
-{# declare a streaming interface #}
-{%- macro fi_streamdecl(size, name) %}
-hls::stream< {{fixed_int(size)}} > {{name}}
-{%- endmacro %}
-
-{# the following macros are defined for Xcelo datatypes #}
-{# declare fixed point number with int_size number of integer bits #}
-{%- macro Xfixed(size, int_size) %}
-fp_int<{{size}},{{int_size}}>
-{%- endmacro %}
-
-{%- macro Xfixed_vdecl(size, int_size, name) %}
-fp_int<{{size}},{{int_size}}> {{name}}
-{%- endmacro %}
-
-{%- macro Xfixed_arr(size, int_size, name, asize) %}
-fp_int<{{size}},{{int_size}}> {{name}}[{{asize}}]
-{%- endmacro %}
-
-{%- macro Xfixed_2darr(size, int_size, name, asize1, asize2) %}
-fp_int<{{size}},{{int_size}}> {{name}}[{{asize1}}][{{asize2}}]
-{%- endmacro %}
-
-{%- macro Xfui(size) %}
-xcelo_uint<{{size}}>
-{%- endmacro %}
-
-{%- macro Xfui_vdecl(size, name) %}
-xcelo_uint<{{size}}> {{name}}
-{%- endmacro %}
-
-{%- macro Xfui_vdecl_arr(size, name, asize) %}
-xcelo_uint<{{size}}> {{name}}[{{asize}}]
-{%- endmacro %}
-
-{%- macro Xfui_arr(size, name, asize) %}
-xcelo_uint<{{size}}> {{name}}[{{asize}}]
-{%- endmacro %}
-
-
diff --git a/DRED_codegen/templates/lrn.inc b/DRED_codegen/templates/lrn.inc
deleted file mode 100644
index 4923ad355575229f4c81fa2eeb24007ecbacce56..0000000000000000000000000000000000000000
--- a/DRED_codegen/templates/lrn.inc
+++ /dev/null
@@ -1,242 +0,0 @@
-{% import "global_macros.inc" as gm %}
-{% import "aux_funcs.inc" as aux %}
-
-{%- macro handle_preprocess(object, name) -%}
- {%- if name in object.keys() %}
- {%- for key, value in object.iteritems() %}
- {%- if key == name %}
- {%- if value > 0 %}
- {%- endif %}
- {%- endif %}
- {%- endfor %}
- {%- endif %}
-{%- endmacro %}
-
-{%- macro fdecl(object) -%}
-{%- endmacro %}
-
-{%- macro var_decl(object) -%}
- {%- if object.preprocess %}
- {%- else %}
- {%- endif %}
-{%- endmacro %}
-
-{% macro fvalid_call(object) %}
-{{gm.ident(1)}}{{object.name}}_test_pool();
-{{gm.ident(1)}}{{object.name}}_test_pool_flatten();
-{% endmacro %}
-
-{# defines paramters passed in when it get called#}
-{%- macro fcall(object, debug) -%}
-{%- set loop_factor = object.loop_factor -%}
-{%- set ip_shift = object.preprocess.ip_shift-%}
-{%- set channels = (object.dshape[2]/loop_factor)|round(0,'ceil')|int-%}
-{{gm.ident(1)}}{{object.name}}_layer ({{object.dshape[0]}}, {{object.dshape[1]}}, {{channels}}, {{object.weights.adjKerNum}}, {{object.weights.K}}, {{object.weights.alpha}}, {{object.weights.beta}},
-{%- if object.interface.in == 'dram' %}
- a,
-{%- elif object.interface.in == 'onchip' %}
- a_onchip,
-{%- endif %}
-{%- if object.interface.int == 'dram' %}
- c
-{%- elif object.interface.int == 'onchip' %}
- c_onchip
-{%- endif %}
-);
-{%- endmacro %}
-
-{% macro readFunc(layer, name, mem, bw, loop_factor)%}
-{%- set iters = (bw*loop_factor/mem)|round(0,'ceil')|int -%}
-int read{{name}}_{{layer}}({{gm.fixed_int(mem)}} *mem,
- {{gm.fixed_int(bw*loop_factor)}} &input,
- int address)
-{
- int index=0;
- //Read requires {{iters}} iterations
-{%for idx in range(0,iters)%}
-{# if this is not the last pack#}
-{%if ((idx+1)*mem < bw*loop_factor) %}
- input.{{gm.pr_range_bound((idx+1)*mem-1,idx*mem)}}=mem[address+index].{{gm.pr_range_bound((mem-1),0)}};
- index ++;
-{# if this is the last pack#}
-{% else %}
- input.{{gm.pr_range_bound(bw*loop_factor-1,idx*mem)}}=mem[address+index].{{gm.pr_range_bound(bw*loop_factor-idx*mem-1,0)}};
-{%endif%}
-{%endfor%}
- return address;
-}
-{% endmacro %}
-
-{% macro writeFunc(layer, name, mem, bw, loop_factor)%}
-{%- set iters = (bw*loop_factor/mem)|round(0,'ceil')|int -%}
-int write{{name}}_{{layer}}({{gm.fixed_int(mem)}} *mem,
- {{gm.fixed_int(bw*loop_factor)}} output,
- int address)
-{
- int index=0;
- //Write requires {{iters}} iterations
-{%for idx in range(0,iters)%}
-{# if this is not the last pack#}
-{%if ((idx+1)*mem < bw*loop_factor) %}
- mem[address+index].{{gm.pr_range_bound((mem-1),0)}} = output.{{gm.pr_range_bound((idx+1)*mem-1,idx*mem)}};
- index ++;
-{# if this is the last pack#}
-{% else %}
- mem[address+index].{{gm.pr_range_bound(bw*loop_factor-idx*mem-1,0)}}=output.{{gm.pr_range_bound(bw*loop_factor-1,idx*mem)}};
-{%endif%}
-{%endfor%}
- return address;
-}
-{% endmacro %}
-
-{%- macro fimpl(object) -%}
-{%- set ip_bw = object.preprocess.ip_bit_width-%}
-{%- set ip_shift = object.preprocess.ip_shift-%}
-{%- set op_bw = object.preprocess.op_bit_width-%}
-{%- set op_shift = object.preprocess.op_shift -%}
-{%- set loop_factor = object.loop_factor -%}
-{%- set mem_bw = object.interface.memory_bit_width -%}
-{%- set channels = (object.dshape[2]/loop_factor)|round(0,'ceil')|int-%}
-{%- set channel_buf_size = 512 -%}
-{# internal bw is the "safe" bit width for the result of multiplication #}
-{%- set iters_input = (ip_bw*loop_factor/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_output = (op_bw*loop_factor/mem_bw)|round(0,'ceil')|int -%}
-
-{{readFunc(object.name, "Input", mem_bw, ip_bw, loop_factor)}}
-{{writeFunc(object.name, "Output", mem_bw, op_bw, loop_factor)}}
-
-void {{object.name}}_read_input_channel(
-{{gm.ident(1)}}int IH,
-{{gm.ident(1)}}int IW,
-{{gm.ident(1)}}int IC,
-{{gm.ident(1)}}ap_int<{{mem_bw}}> *mem,
-{{gm.ident(1)}}{{gm.f_vdecl_arr(ip_bw, ip_bw, 'buffer', channel_buf_size)}},
-{{gm.ident(1)}}int address)
-{
-{{gm.ident(1)}}ap_int<{{ip_bw*loop_factor}}> inPackTemp;
-{% call gm.forloop(1, 'c', 'IC') %}
-{{gm.ident(2)}}readInput_{{object.name}}(mem, inPackTemp, address);
-{% call gm.forloop(2, 'ci', loop_factor) %}
-{{gm.ident(3)}}buffer[c*{{loop_factor}}+ci].{{gm.pr_range(ip_bw)}} = inPackTemp.{{gm.pr_range_idx(ip_bw, 'ci')}};
-{%- endcall -%}
-{{gm.ident(2)}}address += IH*IW*{{iters_input}};
-{%- endcall -%}
-}
-
-void {{object.name}}_write_output_channel(
-{{gm.ident(1)}}int IH,
-{{gm.ident(1)}}int IW,
-{{gm.ident(1)}}int IC,
-{{gm.ident(1)}}ap_int<{{mem_bw}}> *mem,
-{{gm.ident(1)}}{{gm.f_vdecl_arr(ip_bw, ip_bw, 'buffer', channel_buf_size)}},
-{{gm.ident(1)}}int address)
-{
-{{gm.ident(1)}}ap_int<{{ip_bw*loop_factor}}> outPackTemp;
-{% call gm.forloop(1, 'c', 'IC') %}
-{% call gm.forloop(2, 'ci', loop_factor) %}
-{{gm.ident(3)}}outPackTemp.{{gm.pr_range_idx(ip_bw, 'ci')}} = buffer[c*{{loop_factor}}+ci].{{gm.pr_range(ip_bw)}};
-{%- endcall -%}
-{{gm.ident(2)}}writeOutput_{{object.name}}(mem, outPackTemp, address);
-{{gm.ident(2)}}address += IH*IW*{{iters_output}};
-{%- endcall -%}
-}
-
-void {{object.name}}_kernel (
-{{gm.ident(1)}}{{gm.f_vdecl_arr(ip_bw, ip_bw, 'in', channel_buf_size)}},
-{{gm.ident(1)}}{{gm.f_vdecl_arr(ip_bw, ip_bw, 'out', channel_buf_size)}},
-{{gm.ident(1)}}{{gm.f_vdecl(ip_bw, ip_bw, 'kernelNum')}},
-{{gm.ident(1)}}{{gm.f_vdecl(ip_bw, ip_bw, 'adjKerNum')}},
-{{gm.ident(1)}}{{gm.f_vdecl(ip_bw, ip_bw, 'K')}},
-{{gm.ident(1)}}{{gm.f_vdecl(ip_bw, ip_bw, 'alpha')}},
-{{gm.ident(1)}}{{gm.f_vdecl(ip_bw, ip_bw, 'beta')}})
-{
-{{gm.ident(1)}}{{gm.f_vdecl(ip_bw, ip_bw, 'zero')}} = 0;
-{{gm.ident(1)}}{{gm.f_vdecl(ip_bw, ip_bw, 'calKernel')}} = kernelNum -1;
-
-{% call gm.forloop(1, 'i', 'kernelNum') %}
-{{gm.ident(2)}}{{gm.f_vdecl(ip_bw, ip_bw, 'comp1')}} = i-(adjKerNum/2);
-{{gm.ident(2)}}{{gm.f_vdecl(ip_bw, ip_bw, 'comp2')}} = i+(adjKerNum/2);
-{{gm.ident(2)}}{{gm.f_vdecl(ip_bw, ip_bw, 'lowerBound')}} = (comp1 > zero) ? (comp1) : zero;
-{{gm.ident(2)}}{{gm.f_vdecl(ip_bw, ip_bw, 'upperBound')}} = (calKernel < comp2) ? (calKernel) : comp2;
-{{gm.ident(2)}}{{gm.f_vdecl(ip_bw*2, ip_bw*2, 'power')}};
-{{gm.ident(2)}}{{gm.f_vdecl(ip_bw*3, ip_bw*3, 'sum')}} = 0;
-{{gm.ident(2)}}{{gm.f_vdecl(ip_bw*3, ip_bw*3, 'denominator')}};
-{{gm.ident(2)}}{{gm.f_vdecl(ip_bw*3, ip_bw*3, 'scaledSum')}};
-{% call gm.forloop_i(2, 'lowerBound', 'j', '=upperBound', 1)%}
-{{gm.ident(3)}}sum += in[j]*in[j];
-{%- endcall -%}
-{{gm.ident(2)}}scaledSum = K + (alpha * sum);
-{{gm.ident(2)}}//TODO: the data range of this temp should be determined by analyzing the data range
-{{gm.ident(2)}}{{gm.f_vdecl(ip_bw*3, ip_bw, 'temp')}};
-{{gm.ident(2)}}temp.{{gm.pr_range(ip_bw*3)}} = scaledSum.{{gm.pr_range(ip_bw*3)}};
-{{gm.ident(2)}}float tempfloat = temp;
-{{gm.ident(2)}}denominator = pow(tempfloat, (double)beta); //TODO: directly use beta?????
-{{gm.ident(2)}}{{gm.f_vdecl(ip_bw, ip_bw, 'shift_norm_output')}};
-{{gm.ident(2)}}shift_norm_output = in[i]/denominator;
-{{gm.ident(2)}}out = shift_norm_output[i] >> 8;
-{%- endcall -%}
-}
-
-void {{object.name}}_layer(
-{{gm.ident(1)}}int IH,
-{{gm.ident(1)}}int IW,
-{{gm.ident(1)}}int IC,
-{{gm.ident(1)}}{{gm.f_vdecl(ip_bw, ip_bw, 'adjKerNum')}},
-{{gm.ident(1)}}{{gm.f_vdecl(ip_bw, ip_bw, 'K')}},
-{{gm.ident(1)}}{{gm.f_vdecl(ip_bw, ip_bw, 'alpha')}},
-{{gm.ident(1)}}{{gm.f_vdecl(ip_bw, ip_bw, 'beta')}},
-{{gm.ident(1)}}ap_int<{{mem_bw}}>input[50176],
-{{gm.ident(1)}}ap_int<{{mem_bw}}>output[50176])
-{
-{{gm.ident(1)}}{{gm.f_vdecl_arr(ip_bw, ip_bw, 'inTemp', channel_buf_size)}};
-{{gm.ident(1)}}{{gm.f_vdecl_arr(ip_bw, ip_bw, 'outTemp', channel_buf_size)}};
-{% call gm.forloop(1, 'i', 'IH*IW') %}
-{{gm.ident(2)}}//loop index i is also the address of first pack of that fibre
-{{gm.ident(2)}}{{object.name}}_read_input_channel(IH, IW, IC, input, inTemp, i);
-{{gm.ident(2)}}{{object.name}}_kernel(inTemp, outTemp, IC*{{loop_factor}}, adjKerNum, K, alpha, beta);
-{{gm.ident(2)}}{{object.name}}_write_output_channel(IH, IW, IC, output, outTemp, i);
-{%- endcall -%}
-}
-{%- endmacro -%}
-
-{%- macro fvalid(object) -%}
-{%- set IC = object.wshape[2] -%}
-{%- set IH = H+2*padsize -%}
-{%- set IW = W+2*padsize -%}
-{%- set ip_bw = object.preprocess.ip_bit_width-%}
-{%- set op_bw = object.preprocess.op_bit_width-%}
-{%- set mem_bw = object.interface.memory_bit_width -%}
-{%- set onchip_bw = object.interface.onchip_bit_width -%}
-{%- set channels = (object.dshape[2]/loop_factor)|round(0,'ceil')|int-%}
-void {{object.name}}_test_LRN() {
-{{gm.ident(1)}}srand(17);
-{{gm.ident(1)}}//test framework
-{{gm.ident(1)}}int* input_buf = new int[{{IH}}*{{IW}}*{{IC}}];
-{{gm.ident(1)}}int* output_buf = new int[{{IH}}*{{IW}}*{{IC}}];
-{% call gm.forloop(1, 'i', IH*IW*IC) %}
-{{gm.ident(2)}}input_buf[i] = rand()%5;
-{%- endcall -%}
-
-{{gm.ident(1)}}pack_input<int>({{IH}}, {{IW}}, {{IC}}, 0, input_buf, a);
-{{gm.ident(1)}}LRN1_layer({{IH}}, {{IW}}, {{channels}}, {{object.weights.adjKerNum}}, {{object.weights.K}}, {{object.weights.alpha}}, {{object.weights.beta}}, a, c);
-{{gm.ident(1)}}lrn_gold({{IH}}, {{IW}}, {{IC}}, {{object.weights.adjKerNum}}, {{object.weights.K}}, {{object.weights.alpha}}, {{object.weights.beta}}, input_buf, output_buf);
-
-{% call gm.forloop(1, 'i', channels) %}
-{% call gm.forloop(2, 'ii', op_loop_factor) %}
-{% call gm.forloop(3, 'j', H) %}
-{% call gm.forloop(4, 'k', W) %}
-{{gm.ident(5)}}{{gm.fi_vdecl(512, 'temp_c')}} = c[(i*{{H}}*{{W}}+j*{{W}}+k)*{{iters_itrm}}];
-{{gm.ident(5)}}{{gm.fi_vdecl(16, 'out_layer')}} = temp_c.{{gm.pr_range_idx(16,'ii')}};
-{{gm.ident(5)}}int out_golden = output_buf[(i*{{op_loop_factor}}+ii)*{{H}}*{{W}}+j*{{W}}+k];
-{{gm.ident(5)}}if (out_layer != out_golden) {
-{{gm.ident(5)}}cout << "[" << (i*{{op_loop_factor}}+ii) << "][" << j << "][" << k << "]";
-{{gm.ident(5)}}cout << "Diff: Layer - " << out_layer << "Golden - " << out_golden << endl;
-{{gm.ident(5)}}}
-{%- endcall -%}
-{{gm.ident(3)}}//cout << endl;
-{%- endcall -%}
-{{gm.ident(2)}}//cout << endl;
-{%- endcall -%}
-{%- endcall -%}
-}
-{%- endmacro %}
\ No newline at end of file
diff --git a/DRED_codegen/templates/max_pool.inc b/DRED_codegen/templates/max_pool.inc
deleted file mode 100644
index 29051212d562e41f122a85c785ff8c1b08c3d3d2..0000000000000000000000000000000000000000
--- a/DRED_codegen/templates/max_pool.inc
+++ /dev/null
@@ -1,897 +0,0 @@
-{# max_pool.inc #}
-{# Jinja2 template for HLS synthesizable C++ code #}
-{# Inspirit IoT, Inc. 2018 #}
-{# Contributers: #}
-{# Mang Yu #}
-{# Kyle Rupnow #}
-{# Xinheng Liu #}
-
-{% import "global_macros.inc" as gm %}
-{% set ident = gm.ident %}
-{% import "aux_funcs.inc" as aux %}
-
-{# Description: Handle preprocess -- NOT USED #}
-{# ------ #}
-{# Params: #}
-{%- macro handle_preprocess(object, name) -%}
- {%- if name in object.keys() %}
- {%- for key, value in object.iteritems() %}
- {%- if key == name %}
- {%- if value > 0 %}
- {%- endif %}
- {%- endif %}
- {%- endfor %}
- {%- endif %}
-{%- endmacro %}
-
-{# Description: declare the layer functions--NOT USED #}
-{# ------ #}
-{# Params: #}
-{%- macro fdecl(object) -%}
-{%- endmacro %}
-
-{# Description: declare global variabels -- NOT USED #}
-{# ------ #}
-{# Params: #}
-{%- macro var_decl(object) -%}
- {%- if object.preprocess %}
- {%- else %}
- {%- endif %}
-{%- endmacro %}
-
-{# Description: call validation function #}
-{# ------ #}
-{# Params: #}
-{# object -- layer object #}
-{% macro fvalid_call(object) %}
-{% if object.flatten %}
-{{ident(1)}}{{object.name}}_test_pool_flatten();
-{% else %}
-{{ident(1)}}{{object.name}}_test_pool();
-{% endif %}
-{% endmacro %}
-
-{# Description: call layer and activation functions #}
-{# ------ #}
-{# Params: #}
-{# object -- layer object #}
-{# flags -- flags for debugging, verifcation mode, etc. #}
-{%- macro fcall(object, flags) -%}
-{%- set loop_factor = object.loop_factor -%}
-{%- set stride = object.weights.stride[1] -%}
-{%- set ksize = object.weights.ksize[1] -%}
-{%- set padsize = object.weights.padsize -%}
-{%- set mode = object.mode -%}
-{%- set ip_shift = object.preprocess.ip_shift-%}
-{%- set op_shift = object.preprocess.op_shift -%}
-{%- set b_shift = object.weights.b_shift-%}
-{%- set right_shift = ip_shift - op_shift -%}
-{%- set left_shift = ip_shift - b_shift-%}
-{%- set channels = (object.dshape[2]/loop_factor)|round(0,'ceil')|int-%}{# #packets depth-wise#}
-{# read biases from file and put into memory if in verification mode #}
-{% if flags.verify %}
-{{ident(1)}}{{aux.fcall_readb(object)}}
-{% endif %}
-{% if not object.flatten == true %}
-{# decide the size of output depending on type of padding #}
-{# for padding strategy refer to: "https://www.tensorflow.org/api_guides/python/nn#Convolution" #}
-{# TODO TODO TODO: the behavior differs from TensorFlow (8/16/2018) #}
-{% if object.mode == 'SAME' %}
-{%- set OH = (object.dshape[0]/stride+2*padsize)|round(0,'ceil')|int -%}
-{%- set OW = (object.dshape[1]/stride+2*padsize)|round(0,'ceil')|int -%}
-{% else %}
-{%- set OH = ((object.dshape[0]-ksize+1)/stride+2*padsize)|round(0,'ceil')|int -%}
-{%- set OW = ((object.dshape[1]-ksize+1)/stride+2*padsize)|round(0,'ceil')|int -%}
-{% endif %}
-{{ident(1)}}{{object.name}}_bias_relu_pool(
-{%- if object.weights["padding"] == 'SAME' %}
-{{object.weights.padsize}},
-{%- else %}
- 0,
-{%- endif %}
-{{object.dshape[0]}}, {{object.dshape[1]}}, {{channels}}, {{OH}}, {{OW}}, {{left_shift}}, {{right_shift}}, w,
-{# placeholder, in case we need to be able to switch between offchip memory and onchip memory #}
-{%- if object.interface.in == 'dram' %}
- c,
-{%- elif object.interface.in == 'onchip' %}
- c_onchip,
-{%- endif %}
-{%- if object.interface.int == 'dram' %}
- a
-{%- elif object.interface.int == 'onchip' %}
- a_onchip
-{%- endif %}
-);
-{% else %}
-{# if flatten version is required #}
-{% if object.mode == 'SAME' %}
-{%- set OH = (object.dshape[0]/stride)|round(0,'ceil')|int -%}
-{%- set OW = (object.dshape[1]/stride)|round(0,'ceil')|int -%}
-{% else %}
-{%- set OH = ((object.dshape[0]-ksize+1)/stride)|round(0,'ceil')|int -%}
-{%- set OW = ((object.dshape[1]-ksize+1)/stride)|round(0,'ceil')|int -%}
-{% endif %}
-{{ident(1)}}{{object.name}}_bias_relu_pool_flatten({{object.dshape[0]}}, {{object.dshape[1]}}, {{channels}}, {{OH}}, {{OW}}, {{left_shift}}, {{right_shift}}, w,
-{%- if object.interface.in == 'dram' %}
- c,
-{%- elif object.interface.in == 'onchip' %}
- c_onchip,
-{%- endif %}
-{%- if object.interface.int == 'dram' %}
- a
-{%- elif object.interface.int == 'onchip' %}
- a_onchip
-{%- endif %}
-);
-{% endif %}
-{%- endmacro %}
-
-{# Description: implement the memory read & unpack function #}
-{# ------ #}
-{# Params: #}
-{# layer -- name of the layer #}
-{# name -- name of the buffer to read: output/itrm,etc. #}
-{# mem_bw -- bit width of a single memory line #}
-{# bw -- bit width of the data we want to read #}
-{# loop_factor -- loop_factor, also the pack size #}
-{# TODO: share instances of read function between layers #}
-{% macro readFunc(layer, name, mem_bw, bw, loop_factor)%}
-void {{layer}}_read_{{name}}(
- {{gm.Xfui(mem_bw)}} *mem,
- {{gm.Xfixed_arr(bw-1, 0, 'inputToLayer', loop_factor)}},
- int address) {
-{% if mem_bw >= (loop_factor*bw) %}
-{% call gm.forloop(1, 'i', loop_factor) %}
-{%- if bw == 8-%}
-{{ident(2)}}inputToLayer[i]={{gm.Xfixed(bw-1,0)}}::createRaw(mem[address].getChar(i));
-{%- elif bw == 16 -%}
-{{ident(2)}}inputToLayer[i]={{gm.Xfixed(bw-1,0)}}::createRaw(mem[address].getShort(i));
-{%- endif -%}
-{% endcall %}
-{%- else -%}
-{% set mem_num = (mem_bw/bw)|int %}
-{% call gm.forloop_s(1, 'i', loop_factor, mem_num) %}
-{% call gm.forloop(2, 'j', mem_num) %}
-{%- if bw == 8-%}
-{{ident(3)}}inputToLayer[i+j]={{gm.Xfixed(bw-1,0)}}::createRaw(mem[address].getChar(i));
-{%- elif bw == 16 -%}
-{{ident(3)}}inputToLayer[i+j]={{gm.Xfixed(bw-1,0)}}::createRaw(mem[address].getShort(i));
-{%- endif -%}
-{%- endcall -%}
-{%- endcall -%}
-{%- endif -%}
-}
-{% endmacro %}
-
-{# Description: implement the memory write & pack function #}
-{# ------ #}
-{# Params: #}
-{# layer -- name of the layer #}
-{# name -- name of the buffer to write: output/itrm,etc.#}
-{# mem_bw -- bit width of a single memory line #}
-{# bw -- bit width of the data we want to read #}
-{# loop_factor -- loop_factor, also the pack size #}
-{# TODO: share instances of read function between layers #}
-{% macro writeFunc(layer, name, mem_bw, bw, loop_factor)%}
-void {{layer}}_write_{{name}}(
- {{gm.Xfui(mem_bw)}} *mem,
- {{gm.Xfixed_arr(bw-1, 0, 'outputFromLayer', loop_factor)}},
- int address) {
-{% if mem_bw >= (loop_factor*bw) %}
-{% call gm.forloop(1, 'i', loop_factor) %}
-{% if bw == 16 %}
-{{ident(2)}}mem[address].setValue(i, outputFromLayer[i].getRaw());
-{% elif bw == 8 %}
-{{ident(2)}}mem[address].setValue(i, (char)outputFromLayer[i].getRaw());
-{% endif %}
-{% endcall %}
-{%- else -%}
-{% set mem_num = (mem_bw/bw)|int %}
-{% call gm.forloop_s(1, 'i', loop_factor, mem_num) %}
-{% call gm.forloop(2, 'j', mem_num) %}
-{% if bw == 16 %}
-{{ident(3)}}mem[address].setValue(j, outputFromLayer[i+j].getRaw());
-{% elif bw == 8 %}
-{{ident(3)}}mem[address].setValue(j, (char)outputFromLayer[i+j].getRaw());
-{% endif %}
-{%- endcall -%}
-{%- endcall -%}
-{%- endif -%}
-}
-{% endmacro %}
-
-{# Description: implement the layer functions #}
-{# #}
-{# ------ #}
-{# Params: #}
-{# object: layer object that contains all the info #}
-{%- macro fimpl(object) -%}
-{%- set ip_bw = object.preprocess.ip_bit_width-%}
-{%- set ip_shift = object.preprocess.ip_shift-%}
-{%- set op_bw = object.preprocess.op_bit_width-%}
-{%- set op_shift = object.preprocess.op_shift -%}
-{%- set b_bw = object.weights.b_width-%}
-{%- set b_shift = object.weights.b_shift-%}
-{%- set loop_factor = object.loop_factor -%}
-{%- set mem_bw = object.memory_bit_width -%}
-{%- set stride = object.weights.stride[1] -%}
-{%- set ksize = object.weights.ksize[1] -%}
-{%- set mode = object.mode -%}
-{# internal bw is the "safe" bit width for the result of multiplication #}
-{%- set iters_bias = (b_bw*loop_factor/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_input = (ip_bw*loop_factor/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_output = (op_bw*loop_factor/mem_bw)|round(0,'ceil')|int -%}
-
-{# implement memory read/write functions #}
-{{readFunc(object.name, "input", mem_bw, ip_bw, loop_factor)}}
-{{readFunc(object.name, "bias", mem_bw, b_bw, loop_factor)}}
-{{writeFunc(object.name, "output", mem_bw, op_bw, loop_factor)}}
-
-//Implementing {{mode}} mode
-//size {{ksize}}x{{ksize}} stride {{stride}}
-
-/*
-Name: {{object.name}}_saturate
-Description: performs saturation when converting from int_bw to op_bw
-Params:
- in -- number to be converted to output bit-width
-Return:
- temp -- result after saturation in output bit-width
-*/
-{{gm.Xfixed(op_bw-1,0)}} {{object.name}}_saturate(
-{{ident(1)}}{{gm.Xfixed_vdecl(ip_bw-1, 0, 'in')}})
-{
-{{ident(1)}}float OVERFLOWMAX = {{2.0**(op_bw - 1) - 1.0}};
-{{ident(1)}}float OVERFLOWMIN = -{{2.0**(op_bw - 1)}};
-
-{{ident(1)}}{{gm.Xfixed_vdecl(op_bw-1, 0, 'temp')}};
-{{ident(1)}}if(in >= OVERFLOWMAX){
-{{ident(2)}}temp = OVERFLOWMAX;
-{{ident(1)}}} else if (in <= OVERFLOWMIN) {
-{{ident(2)}}temp = OVERFLOWMIN;
-{{ident(1)}}} else {
-{{ident(2)}}temp=in;
-{{ident(1)}}}
-{{ident(1)}}return temp;
-}
-
-/*
-Name: {{object.name}}_write_clear
-Description: write 0 to the memory address specified
-Params:
- mem -- memory to write to
- address -- address that is to be cleared
-*/
-void {{object.name}}_write_clear(
- {{gm.Xfui(mem_bw)}} *mem,
- int address) {
-{{ident(1)}}{{gm.Xfui(mem_bw)}} temp((unsigned int) 0);
-{% if mem_bw >= (loop_factor*op_bw) %}
-{{ident(2)}}mem[address] = temp;
-{%- else -%}
-{% set mem_num = (mem_bw/op_bw)|int %}
-{% call gm.forloop_s(1, 'i', loop_factor, mem_num) %}
-{{ident(2)}}mem[address+i] = temp;
-{%- endcall -%}
-{%- endif -%}
-}
-
-/*
-Name: {{object.name}}_saturate_add
-Description: this function performs add with a saturation,
- it checks the result of addition and clamp the
- output to MAX/MIN value of the datatype
-Params: a -- input value a
- b -- input value b
-Return: c -- output value c after saturation
-TODO: share instances of this between layers??
-*/
-{{gm.Xfixed(ip_bw-1,0)}}{{object.name}}_saturate_add(
-{{gm.Xfixed_vdecl(ip_bw-1, 0, 'a')}},
-{{gm.Xfixed_vdecl(ip_bw-1, 0, 'b')}})
-{
-{{ident(1)}}float OVERFLOWMAX = {{2.0**(ip_bw - 1) - 1.0}};
-{{ident(1)}}float OVERFLOWMIN = -{{2.0**(ip_bw - 1)}};
-
-{{ident(1)}}{{gm.Xfixed_vdecl(ip_bw-1, 0, 'c')}};
-{{ident(1)}}c = a + b;
-{{ident(1)}}if((a > 0)&&(b > 0)&&(c < 0)){
-{{ident(2)}}c = OVERFLOWMAX;
-{{ident(1)}}} else if ((a < 0)&&(b < 0)&&(c > 0)) {
-{{ident(2)}}c = OVERFLOWMIN;
-{{ident(1)}}}
-{{ident(1)}}return c;
-}
-
-{# NOTE: since 2*2 stride 2 is the most common setting for max pooling #}
-{# we have a optimized version for it #}
-{% if (ksize == 2) and (stride == 2) %}
-//NOTE: 2*2 stride 2 version is selected
-/*
-Name: {{object.name}}_maxpool4
-Description: performs max pooling for a single 2*2 window
-Params:
- lat -- data from the input window
- bias -- pack of biases
- rst -- result array
- left_shift -- left shift needed for the bias to align with
- pre-activation
- right_shift -- right shift needed for the pre-activation to
- align wiht the output
-*/
-void {{object.name}}_maxpool4(
- {{gm.Xfixed_2darr(ip_bw-1, 0, 'lat', 4, loop_factor)}},
- {{gm.Xfixed_arr(b_bw-1, 0, 'bias', loop_factor)}},
- {{gm.Xfixed_arr(op_bw-1, 0, 'rst', loop_factor)}},
- int left_shift, int right_shift
-) {
-{% call gm.forloop(1,'j',loop_factor) %}
-{{ident(2)}}#pragma HLS unroll
-{{ident(2)}}{{gm.Xfixed_vdecl(ip_bw-1, 0, 'tempBias')}};
-{{ident(2)}}//unpack bias
-{{ident(2)}}tempBias = bias[j];
-{{ident(2)}}{{gm.Xfixed_vdecl(ip_bw-1, 0, 'a')}};
-{{ident(2)}}{{gm.Xfixed_vdecl(ip_bw-1, 0, 'b')}};
-{{ident(2)}}{{gm.Xfixed_vdecl(ip_bw-1, 0, 'c')}};
-{{ident(2)}}{{gm.Xfixed_vdecl(ip_bw-1, 0, 'd')}};
-{{ident(2)}}//TODO: use _saturate_add() to prevent overflow
-{{ident(2)}}a={{object.name}}_saturate_add(lat[0][j], (tempBias << left_shift))
-{{ident(2)}}a= a >> right_shift;
-{{ident(2)}}b={{object.name}}_saturate_add(lat[1][j], (tempBias << left_shift))
-{{ident(2)}}b= b >> right_shift;
-{{ident(2)}}c={{object.name}}_saturate_add(lat[2][j], (tempBias << left_shift))
-{{ident(2)}}c= c >> right_shift;
-{{ident(2)}}d={{object.name}}_saturate_add(lat[3][j], (tempBias << left_shift))
-{{ident(2)}}d= d >> right_shift;
-{{ident(2)}}{{gm.Xfixed_vdecl(ip_bw-1, 0, 'x')}};
-{{ident(2)}}{{gm.Xfixed_vdecl(ip_bw-1, 0,'y')}};
-{{ident(2)}}{{gm.Xfixed_vdecl(ip_bw-1, 0,'z')}};
-{{ident(2)}}{{gm.Xfixed_vdecl(ip_bw-1, 0,'s')}};
-{{ident(2)}}//perform comparison
-{{ident(2)}}x=a>b?a:b;
-{{ident(2)}}y=c>d?c:d;
-{{ident(2)}}z=x>y?x:y;
-{{ident(2)}}s = z;
-{{ident(2)}}//perform ReLU activation
-{{ident(2)}}if(s > 0.0)
-{{ident(3)}}rst[j] = {{object.name}}_saturate(s);
-{{ident(2)}}else
-{{ident(3)}}rst[j] = 0;
-{% endcall %}
-}
-
-{# flatten and non-flatten version of maxpool are separate #}
-/*
-Name: {{object.name}}_bias_relu_pool
-Description: top level function of max pooling layer. this
- layer performs 3 actions, 1- adding bias, 2-
- max pooling, 3- ReLU activation.
-{% if not object.flatten %}
- This implementation performs 2*2 max poolling,
- wiht stride 2, and will pad the output for the
- next layer according to the padsize parameter.
-Params: See below
-*/
-void {{object.name}}_bias_relu_pool(
- int padsize,
- int IH, //Input height
- int IW, //Input width
- int COO, //number of packed channels
- int OH, //Output height
- int OW, //Output width
- int left_shift, //left shift needed to align bias with the pre-activation
- int right_shift, //right shift needed to align pre-activation with output
- {{gm.Xfui_arr(mem_bw, 'weight', 50176)}},
- {{gm.Xfui_arr(mem_bw, 'in', 50176)}},
- {{gm.Xfui_arr(mem_bw,'out', 50176)}}
-) {
-
- {{gm.Xfixed_arr(b_bw-1, 0, 'bias', loop_factor)}};
- int address = 0;
- int weight_idx = 0;
-{% call gm.forloop(1, 'c', 'COO') %}
-{{ident(2)}}//bias = weight[weight_idx++];
-{{ident(2)}}{{object.name}}_read_bias(weight, bias, weight_idx*{{iters_bias}});
-{{ident(2)}}weight_idx ++;
-{{ident(2)}}//pad leading zero rows
-{% call gm.forloop(2, 'i', 'padsize*OW') %}
-{{ident(3)}}address = (c*OH*OW+i)*{{iters_output}};
-{{ident(3)}}{{object.name}}_write_clear(out, address);
-{% endcall %}
-{% call gm.forloop_s(2, 'h', 'IH', 2) %}
-{{ident(3)}}// pad zeros in front of each line
-{% call gm.forloop(3, 'i', 'padsize') %}
-{{ident(4)}}address = (c*OH*OW+(padsize+h/2)*OW+i)*{{iters_output}};
-{{ident(3)}}{{object.name}}_write_clear(out, address);
-{% endcall %}
-{{ident(3)}}// perform relu/bias/pooling
-{% call gm.forloop_s(3, 'w', 'IW', 2) %}
-{{ident(4)}}{{gm.Xfixed_2darr(ip_bw-1, 0, 'inTemp', 4, loop_factor)}};
-{{ident(4)}}{{gm.Xfixed_arr(op_bw-1, 0, 'outTemp', loop_factor)}};
-{{ident(4)}}//read data from the window
-{{ident(4)}}address = (c*IW*IH+h*IW+w)*{{iters_input}};
-{{ident(4)}}{{object.name}}_read_input(in, inTemp[0], address);
-{{ident(4)}}address = (c*IW*IH+h*IW+w+1)*{{iters_input}};
-{{ident(4)}}{{object.name}}_read_input(in, inTemp[1], address);
-{{ident(4)}}address = (c*IW*IH+(h+1)*IW+w)*{{iters_input}};
-{{ident(4)}}{{object.name}}_read_input(in, inTemp[2], address);
-{{ident(4)}}address = (c*IW*IH+(h+1)*IW+w+1)*{{iters_input}};
-{{ident(4)}}{{object.name}}_read_input(in, inTemp[3], address);
-{{ident(4)}}//perform max pooling
-{{ident(4)}}pool1_maxpool4(inTemp, bias, outTemp, left_shift, right_shift);
-{{ident(4)}}address = (c*OH*OW+(padsize+h/2)*OW+w/2+padsize)*{{iters_output}};
-{{ident(4)}}{{object.name}}_write_output(out, outTemp, address);
-{% endcall %}
-{{ident(3)}}//pad zeros at the end of each line
-{% call gm.forloop(3, 'i', 'padsize') %}
-{{ident(4)}}address = (c*OH*OW+(padsize+h/2)*OW+IW/2+padsize+i)*{{iters_output}};
-{{ident(3)}}{{object.name}}_write_clear(out, address);
-{% endcall %}
-{% endcall %}
-
-{% call gm.forloop(2, 'i', 'padsize*IW') %}
-{{ident(3)}}address = (c*OH*OW+(padsize+IH/2)*OW+i);
-{{ident(3)}}{{object.name}}_write_clear(out, address);
-{% endcall %}
-{% endcall %}
-}
-{% else %}
- This implementation performs 2*2 max poolling,
- wiht stride 2. It will not pad the output for
- the next layer.
-Params: See below
-*/
-void {{object.name}}_bias_relu_pool_flatten(
- int IH, //Input height
- int IW, //Input width
- int COO, //number of packed channels
- int OH, //Output height
- int OW, //Output width
- int left_shift, //left shift needed to align bias with the pre-activation
- int right_shift, //right shift needed to align pre-activation with output
- {# TODO: placeholder for using onchip memory #}
- {{gm.Xfui_arr(mem_bw, 'weight', 50176)}},
- {{gm.Xfui_arr(mem_bw, 'in', 50176)}},
- {{gm.Xfui_arr(mem_bw,'out', 50176)}}
-) {
- {{gm.Xfixed_arr(b_bw-1, 0, 'bias', loop_factor)}};
- int weight_idx = 0;
- int address = 0;
-{% call gm.forloop(1, 'c', 'COO') %}
-{{ident(2)}}{{object.name}}_read_bias(weight, bias, weight_idx*{{iters_bias}});
-{{ident(2)}}weight_idx ++;
-{% call gm.forloop_s(2, 'h', 'IH', 2) %}
-{% call gm.forloop_s(3, 'w', 'IW', 2) %}
-{{ident(4)}}//read in data for the pooling window
-{{ident(4)}}{{gm.Xfixed_2darr(ip_bw-1, 0, 'inTemp', 4, loop_factor)}};
-{{ident(4)}}{{gm.Xfixed_arr(op_bw-1, 0, 'outTemp', loop_factor)}};
-{{ident(4)}}address = (c*IW*IH+h*IW+w)*{{iters_input}};
-{{ident(4)}}{{object.name}}_read_input(in, inTemp[0], address);
-{{ident(4)}}address = (c*IW*IH+h*IW+w+1)*{{iters_input}};
-{{ident(4)}}{{object.name}}_read_input(in, inTemp[1], address);
-{{ident(4)}}address = (c*IW*IH+(h+1)*IW+w)*{{iters_input}};
-{{ident(4)}}{{object.name}}_read_input(in, inTemp[2], address);
-{{ident(4)}}address = (c*IW*IH+(h+1)*IW+w+1)*{{iters_input}};
-{{ident(4)}}{{object.name}}_read_input(in, inTemp[3], address);
-{{ident(4)}}//perform max pooling
-{{ident(4)}}pool1_maxpool4(inTemp, bias, outTemp, left_shift, right_shift);
-{{ident(4)}}address = (c*OH*OW+(h/2)*OW+(w/2))*{{iters_output}};
-{{ident(4)}}//write output to the buffer
-{{ident(4)}}{{object.name}}_write_output(out, outTemp, address);
-{% endcall %}
-{% endcall %}
-{% endcall %}
-}
-{% endif %}
-{# less optimized generic implementation #}
-{%- else -%}
-/*
-Name: {{object.name}}_maxpool_{{ksize}}
-Description: performs max pooling for a single 2*2 window
-Params:
- base -- data to compare with
- rival -- data that will replace the on in base if it is larger
- after adding the bias and is >0
- bias -- pack of biases
- left_shift -- left shift needed for the bias to align with
- pre-activation
-*/
-void {{object.name}}_maxpool_{{ksize}}(
- {{gm.Xfixed_arr(ip_bw-1, 0, 'base', loop_factor)}},
- {{gm.Xfixed_arr(ip_bw-1, 0, 'rival', loop_factor)}},
- {{gm.Xfixed_arr(b_bw-1, 0, 'bias', loop_factor)}},
- int left_shift
-) {
-{% call gm.forloop(1,'j',loop_factor) %}
-{{ident(2)}}{{gm.Xfixed_vdecl(ip_bw-1, 0, 'tempbias')}};
-{{ident(2)}}{{gm.Xfixed_vdecl(ip_bw-1, 0, 'a')}};
-{{ident(2)}}{{gm.Xfixed_vdecl(ip_bw-1, 0, 'b')}};
-{{ident(2)}}tempbias = bias[j];
-{{ident(2)}}a = base[j];
-{{ident(2)}}b = {{object.name}}_saturate_add(rival[j], (tempbias << left_shift));
-{{ident(2)}}{{gm.Xfixed_vdecl(ip_bw-1, 0, 'x')}};
-{{ident(2)}}//compare base and rival+bias
-{{ident(2)}}x=a>b?a:b;
-{{ident(2)}}//perform ReLU activation
-{{ident(2)}}if(x>0)
-{{ident(3)}}base[j] = {{object.name}}_saturate(x);
-{{ident(2)}}else
-{{ident(3)}}base[j]=0;
-{% endcall %}
-}
-
-/*
-Name: {{object.name}}_bias_relu_pool
-Description: top level function of max pooling layer. this
- layer performs 3 actions, 1- adding bias, 2-
- max pooling, 3- ReLU activation.
-{# use non-flatten version of maxpooling #}
-{% if not object.flatten %}
- This implementation performs generic max pooling that can take
- any window size, stride and padding strategy. It will pad the
- output according to the padsize.
-Params: See below
-*/
-void {{object.name}}_bias_relu_pool(
- int padsize, //padsize for the next layer
- int IH, //Input height
- int IW, //Input width
- int CO, //number of packed channels
- int OH, //Output height
- int OW, //Output width
- int left_shift, //left shift needed to align bias with the pre-activation
- int right_shift, //right shift needed to align pre-activation with output
- {{gm.Xfui_arr(mem_bw, 'weight', 50176)}},
- {{gm.Xfui_arr(mem_bw, 'in', 50176)}},
- {{gm.Xfui_arr(mem_bw,'out', 50176)}}
-) {
- {{gm.Xfixed_arr(b_bw-1, 0, 'bias', loop_factor)}};
- int address = 0;
- int weight_idx = 0;
-{% call gm.forloop(1, 'c', 'CO') %}
-{{ident(2)}}{{object.name}}_read_bias(weight, bias, weight_idx*{{iters_bias}});
-{{ident(2)}}weight_idx ++;
-{{ident(2)}}//pad leading zero rows
-{% call gm.forloop(2, 'i', 'padsize*OW') %}
-{{ident(3)}}address = (c*OH*OW+i)*{{iters_output}};
-{{ident(3)}}{{object.name}}_write_clear(out, address);
-{% endcall %}
-{# the boundary for VALID and SAME pooling mode is different #}
-{# TODO TODO TODO: match the behavior with TensorFlow #}
-{%- if (mode == 'VALID') -%}
-{% call gm.forloop_s(2, 'h', 'IH-%d'|format(ksize-1), stride) %}
-{{ident(3)}}// pad zeros in front of each line
-{% call gm.forloop(3, 'i', 'padsize') %}
-{{ident(4)}}address = (c*OH*OW+(padsize+h/2)*OW+i)*{{iters_output}};
-{{ident(4)}}{{object.name}}_write_clear(out, address);
-{% endcall %}
-{{ident(3)}}// perform relu/bias/pooling
-{% call gm.forloop_s(3, 'w', 'IW-%d'|format(ksize-1), stride) %}
-{{ident(4)}}{{gm.Xfixed_arr(ip_bw-1, 0, 'base' , loop_factor)}};
-{{ident(4)}}{{gm.Xfixed_arr(ip_bw-1, 0, 'rival', loop_factor)}};
-{# since array initialization doesn't work, we have to use this stupid loop#}
-{% call gm.forloop(4, 'j', loop_factor)%}
-{{ident(5)}}base[j] = 0;
-{% endcall %}
-{% call gm.forloop_i(4, 0, 'hh', ksize, 1) %}
-{% call gm.forloop_i(5, 0, 'ww', ksize, 1) %}
-{{ident(6)}}if (!(((hh+h)<0)||((ww+w)<0)||((hh+h)>= IH)||((ww+w)>= IW))){
-{{ident(7)}}address = (c*IW*IH+(h+hh)*IW+(w+ww))*{{iters_input}};
-{{ident(7)}}{{object.name}}_read_input(in, rival, address);
-{{ident(7)}}{{object.name}}_maxpool_{{ksize}}(base, rival, bias, left_shift);
-{{ident(6)}}}
-{% endcall %}
-{% endcall %}
-{{ident(4)}}//cast the result to output datatype
-{{ident(4)}}{{gm.Xfixed_arr(op_bw-1, 0,'outCast', loop_factor)}};
-{% call gm.forloop(4, 'j', loop_factor)%}
-{{ident(5)}}outCast[j] = {{object.name}}_saturate(base[j] >> right_shift)
-{% endcall %}
-{{ident(4)}}address = (c*OH*OW+(padsize+h/{{stride}})*OW+w/{{stride}}+padsize)*{{iters_output}};
-{{ident(4)}}{{object.name}}_write_output(out, outCast, address);
-{% endcall %}
-{{ident(3)}}//pad zeros at the end of each line
-{% call gm.forloop(3, 'i', 'padsize') %}
-{{ident(4)}}address = (c*OH*OW+(padsize+h/{{stride}})*OW+OW-padsize+i)*{{iters_output}};
-{{ident(4)}}{{object.name}}_write_clear(out, address);
-{% endcall %}
-{% endcall %}
-{%- elif (mode == 'SAME') -%}
-{% call gm.forloop_s(2, 'h', 'IH', stride) %}
-{{ident(3)}}// pad zeros in front of each line
-{% call gm.forloop(3, 'i', 'padsize') %}
-{{ident(4)}}address = (c*OH*OW+(padsize+h/2)*OW+i)*{{iters_output}};//TODO: Optimize
-{{ident(4)}}{{object.name}}_write_clear(out, address);
-{% endcall %}
-{{ident(3)}}// perform relu/bias/pooling
-{% call gm.forloop_s(3, 'w', 'IW', stride) %}
-{{ident(4)}}#pragma HLS pipeline
-{{ident(4)}}{{gm.Xfixed_arr(ip_bw-1, 0, 'base' , loop_factor)}};
-{{ident(4)}}{{gm.Xfixed_arr(ip_bw-1, 0, 'rival', loop_factor)}};
-{# since array initialization doesn't work, we have to use this stupid loop#}
-{% call gm.forloop(4, 'j', loop_factor)%}
-{{ident(5)}}base[j] = 0;
-{% endcall %}
-{{ident(4)}}//boundary checking
-{% call gm.forloop_i(4, -(ksize//2), 'hh', ksize//2+1, 1) %}
-{% call gm.forloop_i(5, -(ksize//2), 'ww', ksize//2+1, 1) %}
-{{ident(6)}}if (!(((hh+h)<0)||((ww+w)<0)||((hh+h)>= IH)||((ww+w)>= IW))){
-{{ident(7)}}address = (c*IW*IH+(h+hh)*IW+(w+ww))*{{iters_input}};
-{{ident(7)}}{{object.name}}_read_input(in, rival, address);
-{{ident(7)}}{{object.name}}_maxpool_{{ksize}}(base, rival, bias, left_shift);
-{{ident(6)}}}
-{% endcall %}
-{% endcall %}
-{{ident(4)}}{{gm.Xfixed_arr(op_bw-1, 0,'outCast', loop_factor)}};
-{% call gm.forloop(4, 'j', loop_factor)%}
-{{ident(5)}}outCast[j] = {{object.name}}_saturate(base[j] >> right_shift);
-{% endcall %}
-{{ident(4)}}address = (c*OH*OW+(padsize+h/{{stride}})*OW+w/{{stride}}+padsize)*{{iters_output}};
-{{ident(4)}}{{object.name}}_write_output(out, outCast, address);
-{% endcall %}
-{{ident(3)}}//pad zeros at the end of each line
-{% call gm.forloop(3, 'i', 'padsize') %}
-{{ident(4)}}address = (c*OH*OW+(padsize+h/{{stride}})*OW+OW-padsize+i)*{{iters_output}};
-{{ident(4)}}{{object.name}}_write_clear(out, address);
-{% endcall %}
-{% endcall %}
-{% endif %}
-{% call gm.forloop(2, 'i', 'padsize*IW') %}
-{{ident(3)}}address = (c*OH*OW+(IH-padsize)*OW+i);
-{{ident(3)}}{{object.name}}_write_clear(out, address);
-{%- endcall %}
-{%- endcall %}
-}
-{% else %}
-{# implement flatten version of max pooling #}
- This implementation performs generic max pooling that can take
- any window size, stride and padding strategy. This function will
- not pad the output for the next layer. The output should be seen
- as flattened.
-Params: See below
-*/
-void {{object.name}}_bias_relu_pool_flatten(
- int padsize, //padsize for the next layer
- int IH, //Input height
- int IW, //Input width
- int CO, //number of packed channels
- int OH, //Output height
- int OW, //Output width
- int left_shift, //left shift needed to align bias with the pre-activation
- int right_shift, //right shift needed to align pre-activation with output
- {{gm.Xfui_arr(mem_bw, 'weight', 50176)}},
- {{gm.Xfui_arr(mem_bw, 'in', 50176)}},
- {{gm.Xfui_arr(mem_bw,'out', 50176)}}
-) {
- {{gm.Xfixed_arr(b_bw-1, 0, 'bias', loop_factor)}};
- int address = 0;
- int weight_idx = 0;
-{% call gm.forloop(1, 'c', 'CO') %}
-{{ident(2)}}{{object.name}}_read_bias(weight, bias, weight_idx*{{iters_bias}});
-{{ident(2)}}weight_idx ++;
-{% if (mode == 'VALID') %}
-{% call gm.forloop_s(2, 'h', 'IH-%d'|format(ksize-1), stride) %}
-{% call gm.forloop_s(3, 'w', 'IW-%d'|format(ksize-1), stride) %}
-{{ident(4)}}{{gm.Xfixed_arr(ip_bw-1, 0, 'base' , loop_factor)}};
-{{ident(4)}}{{gm.Xfixed_arr(ip_bw-1, 0, 'rival', loop_factor)}};
-{# since array initialization doesn't work, we have to use this stupid loop#}
-{% call gm.forloop(4, 'j', loop_factor)%}
-{{ident(5)}}base[j] = 0;
-{% endcall %}
-{% call gm.forloop_i(4, 0, 'hh', ksize, 1) %}
-{% call gm.forloop_i(5, 0, 'ww', ksize, 1) %}
-{{ident(6)}}if (!(((hh+h)<0)||((ww+w)<0)||((hh+h)>= IH)||((ww+w)>= IW))){
-{{ident(7)}}address = (c*IW*IH+(h+hh)*IW+(w+ww))*{{iters_input}};
-{{ident(7)}}{{object.name}}_read_input(in, rival, address);
-{{ident(7)}}{{object.name}}_maxpool_{{ksize}}(base, rival, bias, left_shift);
-{{ident(6)}}}
-{% endcall %}
-{% endcall %}
-{{ident(4)}}{{gm.Xfixed_arr(op_bw-1, 0,'outCast', loop_factor)}};
-{% call gm.forloop(4, 'j', loop_factor)%}
-{{ident(5)}}outCast[j] = {{object.name}}_saturate(base[j] >> right_shift)
-{% endcall %}
-{{ident(4)}}address = (c*OH*OW+(h/{{stride}})*OW+w/{{stride}})*{{iters_output}};
-{{ident(4)}}{{object.name}}_write_output(out, outCast, address);
-{% endcall %}
-{% endcall %}
-{%- elif (mode == 'SAME') -%}
-{% call gm.forloop_s(2, 'h', 'IH', stride) %}
-{% call gm.forloop_s(3, 'w', 'IW', stride) %}
-{{ident(4)}}{{gm.Xfixed_arr(ip_bw-1, 0, 'base' , loop_factor)}};
-{{ident(4)}}{{gm.Xfixed_arr(ip_bw-1, 0, 'rival', loop_factor)}};
-{# since array initialization doesn't work, we have to use this stupid loop#}
-{% call gm.forloop(4, 'j', loop_factor)%}
-{{ident(5)}}base[j] = 0;
-{% endcall %}
-{% call gm.forloop_i(4, -(ksize//2), 'hh', ksize//2+1, 1) %}
-{% call gm.forloop_i(5, -(ksize//2), 'ww', ksize//2+1, 1) %}
-{{ident(6)}}if (!(((hh+h)<0)||((ww+w)<0)||((hh+h)>= IH)||((ww+w)>= IW))){
-{{ident(7)}}address = (c*IW*IH+(h+hh)*IW+(w+ww))*{{iters_input}};
-{{ident(7)}}{{object.name}}_read_input(in, rival, address);
-{{ident(7)}}{{object.name}}_maxpool_{{ksize}}(base, rival, bias, left_shift);
-{{ident(6)}}}
-{% endcall %}
-{% endcall %}
-{{ident(4)}}{{gm.Xfixed_arr(op_bw-1, 0,'outCast', loop_factor)}};
-{% call gm.forloop(4, 'j', loop_factor)%}
-{{ident(5)}}outCast[j] = {{object.name}}_saturate(base[j] >> right_shift)
-{% endcall %}
-{{ident(4)}}address = (c*OH*OW+(h/{{stride}})*OW+w/{{stride}})*{{iters_output}};
-{{ident(4)}}{{object.name}}_write_output(out, outCast, address);
-{% endcall %}
-{% endcall %}
-{% endif %}
-{% endcall %}
-}
-{% endif %}
-{% endif %}
-{%- endmacro %}
-
-{% macro fvalid(object, flags) %}
-{%- set ip_bw = object.preprocess.ip_bit_width-%}
-{%- set ip_shift = object.preprocess.ip_shift-%}
-{%- set op_bw = object.preprocess.op_bit_width-%}
-{%- set op_shift = object.preprocess.op_shift -%}
-{%- set b_bw = object.weights.b_width-%}
-{%- set b_shift = object.weights.b_shift-%}
-{%- set loop_factor = object.loop_factor -%}
-{%- set mem_bw = object.memory_bit_width -%}
-{%- set stride = object.weights.stride[1] -%}
-{%- set ksize = object.weights.ksize[1] -%}{# window size for pooling #}
-{%- set padsize = object.weights.padsize-%}
-{%- set mode = object.mode -%} {# mode of doing pooling #}
-{%- set H = object.dshape[0] -%}
-{%- set W = object.dshape[1] -%}
-{%- set C = object.dshape[2] -%}
-{%- set channels = (object.dshape[2]/loop_factor)|round(0,'ceil')|int-%}
-{# internal bw is the "safe" bit width for the result of multiplication #}
-{%- set iters_bias = (b_bw*loop_factor/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_input = (ip_bw*loop_factor/mem_bw)|round(0,'ceil')|int -%}
-{%- set iters_output = (op_bw*loop_factor/mem_bw)|round(0,'ceil')|int -%}
-
-{% if object.flatten %}
-{% if mode == 'SAME'%}
-{%- set OH = (object.dshape[0]/stride)|round(0,'ceil')|int -%}
-{%- set OW = (object.dshape[1]/stride)|round(0,'ceil')|int -%}
-{% elif mode == 'VALID' %}
-{%- set OH = ((object.dshape[0]-ksize+1)/stride)|round(0,'ceil')|int -%}
-{%- set OW = ((object.dshape[1]-ksize+1)/stride)|round(0,'ceil')|int -%}
-{% endif %}
-void {{object.name}}_test_pool_flatten() {
- srand(17);
- bool flag = false;
- int input_buf[{{H}}*{{W}}*{{C}}] = {0};
- int weight_buf[{{C}}];
- int output_buf[{{OH}}*{{OW}}*{{C}}];
-
-{% call gm.forloop(1, 'i', C) %}
-{% call gm.forloop(2, 'j', H) %}
-{% call gm.forloop(3, 'k', W) %}
-{{ident(4)}}input_buf[i*{{H}}*{{W}}+j*{{W}}+k] = rand()%4 - 2;
-{% endcall %}
-{% endcall %}
-{% endcall %}
-
-{{ident(1)}}pack_input<int>({{H}}, {{W}}, {{C}}, 0, {{loop_factor}}, {{ip_bw}}, input_buf, (ap_int<512>*)a);
-
-{% call gm.forloop(1, 'i', C) %}
-{{ident(2)}}weight_buf[i] = rand()%2;
-{% endcall %}
-
-{% call gm.forloop(1, 'i', (object.dshape[2]/loop_factor)|round(0,'ceil')|int) %}
-{% call gm.forloop(1, 'ii', loop_factor) %}
-{{ident(2)}}if((i*{{loop_factor}}+ii) < {{C}}){
-{{ident(2)}}((ap_int<512>*)w)[i].{{gm.pr_range_idx(b_bw, 'ii')}} = weight_buf[i*{{loop_factor}}+ii];
-{{ident(2)}}} else {
-{{ident(2)}}((ap_int<512>*)w)[i].{{gm.pr_range_idx(b_bw, 'ii')}} = 0;
-{{ident(2)}}}
-{% endcall %}
-{% endcall %}
-
-{% call gm.forloop(1, 'i', 50176) %}
-{{ident(2)}}((ap_int<512>*)c)[i] = 0;
-{% endcall %}
-
-{{ident(1)}}maxpool_gold({{H}}, {{W}}, {{C}}, 0, {{ksize}}, {{stride}},
-{%- if mode == 'SAME' -%}
-true,
-{%- elif mode == 'VALID' -%}
-false,
-{%- endif -%}
- input_buf, weight_buf, output_buf);
-{{ident(1)}}{{object.name}}_bias_relu_pool_flatten({{object.dshape[0]}}, {{object.dshape[1]}}, {{(object.dshape[2]/loop_factor)|round(0,'ceil')|int}}, {{OH}}, {{OW}}, 0, 0, (xcelo_uint<{{mem_bw}}>*)w, (xcelo_uint<{{mem_bw}}>*)a, (xcelo_uint<{{mem_bw}}>*)c);
-{% call gm.forloop(1, 'i', 1) %}
-{% call gm.forloop(2, 'ii', loop_factor) %}
-{% call gm.forloop(3, 'j', OH) %}
-{% call gm.forloop(4, 'k', OW) %}
-{{ident(5)}}{{gm.fi_vdecl(mem_bw, 'temp_c')}} = ((ap_int<512>*)c)[(i*{{OH}}*{{OW}}+j*{{OW}}+k)*{{iters_output}}];
-{{ident(5)}}{{gm.fi_vdecl(op_bw, 'out_layer')}} = temp_c.{{gm.pr_range_idx(op_bw,'ii')}};
-{{ident(5)}}int out_golden = output_buf[(i*{{loop_factor}}+ii)*{{OH}}*{{OW}}+j*{{OW}}+k];
-{{ident(5)}}if (out_layer != out_golden) {
-{{ident(5)}}cout << "[" << (i*{{loop_factor}}+ii) << "][" << j << "][" << k << "]";
-{{ident(5)}}cout << "Diff: Layer - " << out_layer << "Golden - " << out_golden << endl;
-{{ident(5)}}}
-{%- endcall -%}
-{{ident(3)}}//cout << endl;
-{%- endcall -%}
-{{ident(2)}}//cout << endl;
-{%- endcall -%}
-{%- endcall -%}
-{{ident(1)}}assert(!flag);
-}
-{# if not flatten #}
-{% else %}
-{% if mode == 'SAME'%}
-{%- set OH = (object.dshape[0]/stride+2*padsize)|round(0,'ceil')|int -%}
-{%- set OW = (object.dshape[1]/stride+2*padsize)|round(0,'ceil')|int -%}
-{% elif mode == 'VALID' %}
-{%- set OH = ((object.dshape[0]-ksize+1)/stride+2*padsize)|round(0,'ceil')|int -%}
-{%- set OW = ((object.dshape[1]-ksize+1)/stride+2*padsize)|round(0,'ceil')|int -%}
-{% endif %}
-void {{object.name}}_test_pool() {
- srand(17);
- bool flag = false;
- int input_buf[{{H}}*{{W}}*{{C}}] = {0};
- int weight_buf[{{C}}];
- int output_buf[{{OH}}*{{OW}}*{{C}}];
-
-{% call gm.forloop(1, 'i', C) %}
-{% call gm.forloop(2, 'j', H) %}
-{% call gm.forloop(3, 'k', W) %}
-{{ident(4)}}input_buf[i*{{H}}*{{W}}+j*{{W}}+k] = rand()%4 - 2;
-{% endcall %}
-{% endcall %}
-{% endcall %}
-
-{{ident(1)}}pack_input<int>({{H}}, {{W}}, {{C}}, 0, {{loop_factor}}, {{ip_bw}}, input_buf, (ap_int<512>*)a);
-
-{% call gm.forloop(1, 'i', C) %}
-{{ident(2)}}weight_buf[i] = rand()%2;
-{% endcall %}
-
-{% call gm.forloop(1, 'i', (object.dshape[2]/loop_factor)|round(0,'ceil')|int) %}
-{% call gm.forloop(1, 'ii', loop_factor) %}
-{{ident(2)}}if((i*{{loop_factor}}+ii) < {{C}}){
-{{ident(2)}}((ap_int<512>*)w)[i].{{gm.pr_range_idx(b_bw, 'ii')}} = weight_buf[i*{{loop_factor}}+ii];
-{{ident(2)}}} else {
-{{ident(2)}}((ap_int<512>*)w)[i].{{gm.pr_range_idx(b_bw, 'ii')}} = 0;
-{{ident(2)}}}
-{% endcall %}
-{% endcall %}
-
-{% call gm.forloop(1, 'i', 50176) %}
-{{ident(2)}}((ap_int<512>*)c)[i] = 0;
-{% endcall %}
-
-{{ident(1)}}maxpool_gold({{H}}, {{W}}, {{C}}, {{padsize}}, {{ksize}}, {{stride}},
-{%- if mode == 'SAME' -%}
-true,
-{%- elif mode == 'VALID' -%}
-false,
-{%- endif -%}
- input_buf, weight_buf, output_buf);
-{{ident(1)}}{{object.name}}_bias_relu_pool({{padsize}}, {{object.dshape[0]}}, {{object.dshape[1]}}, {{(object.dshape[2]/loop_factor)|round(0,'ceil')|int}}, {{OH}}, {{OW}}, 0, 0, w, a, c);
-
-{% call gm.forloop(1, 'i', 1) %}
-{% call gm.forloop(2, 'ii', loop_factor) %}
-{% call gm.forloop(3, 'j', OH) %}
-{% call gm.forloop(4, 'k', OW) %}
-{{ident(5)}}{{gm.fi_vdecl(512, 'temp_c')}} = ((ap_int<512>*)c)[(i*{{OH}}*{{OW}}+j*{{OW}}+k)*{{iters_output}}];
-{{ident(5)}}{{gm.fi_vdecl(8, 'out_layer')}} = temp_c.{{gm.pr_range_idx(8,'ii')}};
-{{ident(5)}}int out_golden = output_buf[(i*{{loop_factor}}+ii)*{{OH}}*{{OW}}+j*{{OW}}+k];
-{{ident(5)}}if (out_layer != out_golden) {
-{{ident(5)}}cout << "[" << (i*{{loop_factor}}+ii) << "][" << j << "][" << k << "]";
-{{ident(5)}}cout << "Diff: Layer - " << out_layer << "Golden - " << out_golden << endl;
-{{ident(5)}}flag = true;
-{{ident(5)}}}
-{%- endcall -%}
-{{ident(3)}}//cout << endl;
-{%- endcall -%}
-{{ident(2)}}//cout << endl;
-{%- endcall -%}
-{%- endcall -%}
-{{ident(1)}}assert(!flag);
-}
-{% endif %}
-{% endmacro %}
\ No newline at end of file
diff --git a/DRED_codegen/templates/net_main.cpp.jinja b/DRED_codegen/templates/net_main.cpp.jinja
deleted file mode 100644
index 4cda3bd1f3e3d7e76a63b12c1b4251f220a30d94..0000000000000000000000000000000000000000
--- a/DRED_codegen/templates/net_main.cpp.jinja
+++ /dev/null
@@ -1,245 +0,0 @@
-{% import 'global_macros.inc' as gm %}
-{% set ident = gm.ident %}
-{# Description: macro for layer function declarations #}
-{# ------ #}
-{# Params: #}
-{# object -- layer object #}
-{# name -- type name of the layer. e.g. conv,fc #}
-{# template -- filename of the template #}
-{%- macro fdecl_template(object, name, template) -%}
-{% from template import fdecl as lfunc %}
-{%- for layer in object %}
-{%- if layer.layer_type == name %}
-{{lfunc(layer)}}
-{%- endif %}
-{%- endfor %}
-{%- endmacro %}
-
-{# Description: macro for global variable declarations #}
-{# ------ #}
-{# Params: #}
-{# object -- layer object #}
-{# name -- type name of the layer. e.g. conv,fc #}
-{# template -- filename of the template #}
-{%- macro var_decl_template(object, name, template) -%}
-{% from template import var_decl as lfunc %}
-{%- for layer in object %}
-{%- if layer.layer_type == name %}
-{{lfunc(layer)}}
-{%- endif %}
-{%- endfor %}
-{%- endmacro %}
-
-{# Description: macro for layer function implementations #}
-{# ------ #}
-{# Params: #}
-{# object -- layer object #}
-{# name -- type name of the layer. e.g. conv, fc, etc. #}
-{# template -- filename of the template #}
-{%- macro fimpl_template(object, name, template) -%}
-{% from template import fimpl as lfunc %}
-{%- for layer in object %}
-{%- if layer.layer_type == name %}
-{{lfunc(layer)}}
-{%- endif %}
-{%- endfor %}
-{%- endmacro %}
-
-{# Description: implemente the validation func for layers #}
-{# ------ #}
-{# Params: #}
-{# object -- layer object #}
-{# name -- type name of the layer. e.g. conv, fc, etc. #}
-{# template -- filename of the template #}
-{%- macro fvalid_template(object, name, template) -%}
-{% from template import fvalid as lfunc %}
-{%- for layer in object %}
-{%- if layer.layer_type == name %}
-{# to validate FC layer we need info about previous layer #}
-{%- if loop.previtem -%}
-{%- set prev = loop.previtem -%}
-{%- else -%}
-{%- set prev = layer -%} {# temporary hack to avoid error #}
-{%- endif -%}
-{%- if layer.layer_type == 'fc'-%}
-{{lfunc(layer, prev, flags)}}
-{%- else -%}
-{{lfunc(layer, flags)}}
-{%- endif -%}
-{%- endif %}
-{%- endfor %}
-{%- endmacro %}
-
-{# Description: implemente the validation func for layers #}
-{# ------ #}
-{# Params: #}
-{# object -- layer object #}
-{# name -- type name of the layer. e.g. conv, fc, etc. #}
-{# template -- filename of the template #}
-{%- macro fvalid_call_template(object, flags, name, template) -%}
-{% from template import fvalid_call as lfunc %}
-{%- for layer in object %}
-{% if layer.layer_type == name %}
-{{lfunc(layer)}}
-{%- endif %}
-{%- endfor %}
-{%- endmacro %}
-
-{# Description: implemente the validation func for layers #}
-{# ------ #}
-{# Params: #}
-{# object -- layer object #}
-{# name -- type name of the layer. e.g. conv, fc, etc. #}
-{# template -- filename of the template #}
-{%- macro fcall_network(network, flags) -%}
-{% for layer in network %}
-{% if layer.layer_type == 'conv' %}
-{% from 'convolution.inc' import fcall as lfunc %}
-{{lfunc(layer, flags)}}
-{% elif layer.layer_type == 'avg_pool' %}
-{% from 'avg_pool.inc' import fcall as lfunc %}
-{{lfunc(layer, flags)}}
-{% elif layer.layer_type == 'max_pool' %}
-{% from 'max_pool.inc' import fcall as lfunc %}
-{{lfunc(layer, flags)}}
-{% elif layer.layer_type == 'fc' %}
-{% from 'fc.inc' import fcall as lfunc %}
-{# FC layer need info of the previous layer #}
-{{lfunc(layer, loop.previtem, flags)}}
-{% elif layer.layer_type == 'batch_norm' %}
-{% from 'batch_norm.inc' import fcall as lfunc %}
-{{lfunc(layer, flags)}}
-{% elif layer.layer_type == 'lrn' %}
-{% from 'lrn.inc' import fcall as lfunc %}
-{{lfunc(layer, flags)}}
-{% endif %}
-{% endfor %}
-{%- endmacro -%}
-
-#include "xcelo_fp_int.h"
-#include "xcelo_uint.h"
-{# C/C++ headers only needed for C verification and debug #}
-{% if flags.verify %}
-#include <iostream>
-#include <fstream>
-#include <vector>
-#include <sys/resource.h>
-#include <cmath>
-{# aux functions need ap_int to put packed data into memory #}
-#include "ap_int.h"
-{# assertion requred for debugging purpose #}
-{% if flags.debug %}
-#include <assert.h>
-{% endif %}
-{% endif %}
-using namespace std;
-
-{# intended to be offchip memory interface #}
-xcelo_uint<512> a[100352];
-xcelo_uint<512> w[100352];
-xcelo_uint<512> c[100352];
-{# placeholder, reserved for onchip memory interface #}
-xcelo_uint<512> a_onchip[1];
-xcelo_uint<512> c_onchip[1];
-
-{# declare all layer functions, currently not used #}
-//Function declarations
-{{fdecl_template(network, 'conv', 'convolution.inc')}}
-{{fdecl_template(network, 'avg_pool', 'avg_pool.inc')}}
-{{fdecl_template(network, 'max_pool', 'max_pool.inc')}}
-{{fdecl_template(network, 'fc', 'fc.inc')}}
-
-{# declare all global variables for each layer, currently not used #}
-//Variable declarations
-{{var_decl_template(network, 'conv', 'convolution.inc')}}
-{{var_decl_template(network, 'avg_pool', 'avg_pool.inc')}}
-{{var_decl_template(network, 'max_pool', 'max_pool.inc')}}
-{{var_decl_template(network, 'fc', 'fc.inc')}}
-
-{# generate implementations for each type of layer #}
-//Function implementations
-{{fimpl_template(network, 'conv', 'convolution.inc')}}
-{{fimpl_template(network, 'avg_pool', 'avg_pool.inc')}}
-{{fimpl_template(network, 'max_pool', 'max_pool.inc')}}
-{{fimpl_template(network, 'fc', 'fc.inc')}}
-{{fimpl_template(network, 'batch_norm', 'batch_norm.inc')}}
-{{fimpl_template(network, 'lrn', 'lrn.inc')}}
-
-{# instantiate auxiliary functions if doing verification #}
-{% if flags.verify %}
-{# Import all auxiliary functions: Read Weight/Input, Display Array, etc.#}
-{% from 'aux_funcs.inc' import fimpl_aux %}
-{{fimpl_aux()}}
-{# implement validation functions if verifying a layer#}
-{% if flags.verify == 'layer' %}
-//Function validations
-{{fvalid_template(network, 'conv', 'convolution.inc')}}
-{{fvalid_template(network, 'avg_pool', 'avg_pool.inc')}}
-{{fvalid_template(network, 'max_pool', 'max_pool.inc')}}
-{{fvalid_template(network, 'fc', 'fc.inc')}}
-{# TODO: add validation functions for LRN and BN layer#}
-{% endif %}
-{% endif %}
-
-{######################## MAIN FUNCTION IN C++ ###########################}
-int main(int argc, char* argv[]) {
-{# if we are verifying a single layer#}
-{% if flags.verify == 'layer' %}
-{# TODO: wrap the following in a macro #}
-{# TODO: add BN and LRN to the following list of layers#}
-{{fvalid_call_template(network, flags, 'conv', 'convolution.inc')}}
-{{fvalid_call_template(network, flags, 'max_pool', 'max_pool.inc')}}
-{{fvalid_call_template(network, flags, 'fc', 'fc.inc')}}
-{% if flags.debug %}
-{{ident(1)}}cout << "per layer test passed" << endl;
-{% endif %}
-{# if we are verifying a network #}
-{% elif flags.verify == 'network' %}
-{# get shape of input tensor to the network from the first layer #}
-{%set IH = network[0].dshape[0] %}
-{%set IW = network[0].dshape[1] %}
-{%set IC = network[0].dshape[2] %}
-{%set pad = (network[0].wshape[0]-1)/2 %}
-{# NOTE: int8_t in the following line can be changed if input datatype is larger than 8 #}
-{{ident(1)}}read_input<int8_t>(argv[1], {{IH}}, {{IW}}, {{IC}}, {{pad}}, (ap_int<512>*)a);
-
-{# call all layer functions#}
-{{fcall_network(network, flags)}}
-
-{{ident(1)}}//{{flags.test_method}} {# show test method in as a comment in C code#}
-{% if flags.test_method == 'groundtruth' %}
-{# get parameters on logits from the #}
-{# ATTENTION: the following two variables should be changed according to the network #}
-{% set logit_bw = 8 %}
-{% set num_classes = 10 %}
-
-{{ident(1)}}ap_int<{{logit_bw}}> result = 0;
-{{ident(1)}}int classification;
-{% call gm.forloop(1, 'i', num_classes) %}
-{{ident(2)}}ap_int<512> temp_a = ((ap_int<512>*)a)[0];
-{{ident(2)}}ap_int<{{logit_bw}}> t = temp_a.range({{logit_bw}}*i+{{logit_bw-1}},{{logit_bw}}*i);
-{% if flags.debug %}{# display logic for each class if in debugging mode #}
-{{ident(2)}}cout << t << ",";
-{% endif %}
-{# perform simple argmax #}
-{{ident(2)}}if (t > result){
-{{ident(3)}}result = t;
-{{ident(3)}}classification = i;
-{{ident(2)}}}
-{% endcall %}
-{% if flags.debug %}{# display result of classification in debugging mode #}
-{{ident(1)}}//simple argmax, and show result of classification
-{{ident(1)}}cout << "Classification result: " << classification << endl;
-{% endif %}
-{# if verifying network in with golden input & output#}
-{% elif flags.test_method == 'golden' %}
-{{ident(1)}}//compare with golden output
-{{ident(1)}}//TODO TODO TODO
-{% endif %}
-{# else in clean mode #}
-{% else %}
-{# call all layer functions#}
-//In clean mode
-{{fcall_network(network, flags)}}
-{% endif %}
-}
\ No newline at end of file
diff --git a/DRED_codegen/templates/plot.ipynb b/DRED_codegen/templates/plot.ipynb
deleted file mode 100644
index caea6169cc80c1a4649e9582d18401babc8e807a..0000000000000000000000000000000000000000
--- a/DRED_codegen/templates/plot.ipynb
+++ /dev/null
@@ -1,34 +0,0 @@
-{
- "cells": [
- {
- "cell_type": "code",
- "execution_count": null,
- "metadata": {
- "collapsed": true
- },
- "outputs": [],
- "source": []
- }
- ],
- "metadata": {
- "kernelspec": {
- "display_name": "Python 2",
- "language": "python",
- "name": "python2"
- },
- "language_info": {
- "codemirror_mode": {
- "name": "ipython",
- "version": 2
- },
- "file_extension": ".py",
- "mimetype": "text/x-python",
- "name": "python",
- "nbconvert_exporter": "python",
- "pygments_lexer": "ipython2",
- "version": "2.7.6"
- }
- },
- "nbformat": 4,
- "nbformat_minor": 0
-}
diff --git a/DRED_codegen/templates/test.jinja b/DRED_codegen/templates/test.jinja
deleted file mode 100644
index 1ac6b2da0284f255f51b2448ab5759afe594188b..0000000000000000000000000000000000000000
--- a/DRED_codegen/templates/test.jinja
+++ /dev/null
@@ -1 +0,0 @@
-Hello {{world}}!
\ No newline at end of file
diff --git a/DRED_codegen/test logs b/DRED_codegen/test logs
deleted file mode 100644
index 4861d209da31e46eb9b526261629fb91ad86ba02..0000000000000000000000000000000000000000
--- a/DRED_codegen/test logs
+++ /dev/null
@@ -1,7 +0,0 @@
-test logs
-
-state | network | reward | result
-12 | 3 layer | x1 -1000 -1500 | agent stays at 12
-16 | 3 layer | x10 -1000 -1500 |agent stays at 12
-16 | 3 layer | x10 -5 -10 | agent stays at 12
-16 | 4 layer 128-256-128 |x100 -1 -1 | not known
\ No newline at end of file
diff --git a/DRED_codegen/Plot.ipynb b/Plot.ipynb
similarity index 100%
rename from DRED_codegen/Plot.ipynb
rename to Plot.ipynb
diff --git a/DRED_codegen/codegen.py b/codegen.py
similarity index 100%
rename from DRED_codegen/codegen.py
rename to codegen.py
diff --git a/DRED_codegen/conv template base.cpp b/conv template base.cpp
similarity index 100%
rename from DRED_codegen/conv template base.cpp
rename to conv template base.cpp
diff --git a/DRED_codegen/exhaustive_search.py b/exhaustive_search.py
similarity index 100%
rename from DRED_codegen/exhaustive_search.py
rename to exhaustive_search.py
diff --git a/DRED_codegen/graph_xc7a200tfbg676-2_10_10.npy b/graph_xc7a200tfbg676-2_10_10.npy
similarity index 100%
rename from DRED_codegen/graph_xc7a200tfbg676-2_10_10.npy
rename to graph_xc7a200tfbg676-2_10_10.npy
diff --git a/DRED_codegen/graph_xc7a200tfbg676-2_11_11.npy b/graph_xc7a200tfbg676-2_11_11.npy
similarity index 100%
rename from DRED_codegen/graph_xc7a200tfbg676-2_11_11.npy
rename to graph_xc7a200tfbg676-2_11_11.npy
diff --git a/DRED_codegen/graph_xc7a200tfbg676-2_8_8.npy b/graph_xc7a200tfbg676-2_8_8.npy
similarity index 100%
rename from DRED_codegen/graph_xc7a200tfbg676-2_8_8.npy
rename to graph_xc7a200tfbg676-2_8_8.npy
diff --git a/DRED_codegen/graph_xc7a200tfbg676-2_9_9.npy b/graph_xc7a200tfbg676-2_9_9.npy
similarity index 100%
rename from DRED_codegen/graph_xc7a200tfbg676-2_9_9.npy
rename to graph_xc7a200tfbg676-2_9_9.npy
diff --git a/DRED_codegen/graph_xc7vx485tffg1761-2_10_10.npy b/graph_xc7vx485tffg1761-2_10_10.npy
similarity index 100%
rename from DRED_codegen/graph_xc7vx485tffg1761-2_10_10.npy
rename to graph_xc7vx485tffg1761-2_10_10.npy
diff --git a/DRED_codegen/graph_xc7z020clg484-1_10_10.npy b/graph_xc7z020clg484-1_10_10.npy
similarity index 100%
rename from DRED_codegen/graph_xc7z020clg484-1_10_10.npy
rename to graph_xc7z020clg484-1_10_10.npy
diff --git a/DRED_codegen/graph_xc7z020clg484-1_10_10_search.npy b/graph_xc7z020clg484-1_10_10_search.npy
similarity index 100%
rename from DRED_codegen/graph_xc7z020clg484-1_10_10_search.npy
rename to graph_xc7z020clg484-1_10_10_search.npy
diff --git a/DRED_codegen/graph_xc7z020clg484-1_8_8.npy b/graph_xc7z020clg484-1_8_8.npy
similarity index 100%
rename from DRED_codegen/graph_xc7z020clg484-1_8_8.npy
rename to graph_xc7z020clg484-1_8_8.npy
diff --git a/DRED_codegen/graph_xcvu190-flgc2104-2-e_10_10.npy b/graph_xcvu190-flgc2104-2-e_10_10.npy
similarity index 100%
rename from DRED_codegen/graph_xcvu190-flgc2104-2-e_10_10.npy
rename to graph_xcvu190-flgc2104-2-e_10_10.npy
diff --git a/DRED_codegen/graph_xcvu190-flgc2104-2-e_10_10_search.npy b/graph_xcvu190-flgc2104-2-e_10_10_search.npy
similarity index 100%
rename from DRED_codegen/graph_xcvu190-flgc2104-2-e_10_10_search.npy
rename to graph_xcvu190-flgc2104-2-e_10_10_search.npy
diff --git a/DRED_codegen/graph_xcvu9p-flga2104-2L-e-es1_10_10.npy b/graph_xcvu9p-flga2104-2L-e-es1_10_10.npy
similarity index 100%
rename from DRED_codegen/graph_xcvu9p-flga2104-2L-e-es1_10_10.npy
rename to graph_xcvu9p-flga2104-2L-e-es1_10_10.npy
diff --git a/DRED_codegen/history_trace_back.npy b/history_trace_back.npy
similarity index 100%
rename from DRED_codegen/history_trace_back.npy
rename to history_trace_back.npy
diff --git a/DRED_codegen/history_trace_back3.npy b/history_trace_back3.npy
similarity index 100%
rename from DRED_codegen/history_trace_back3.npy
rename to history_trace_back3.npy
diff --git a/DRED_codegen/history_trace_back_v2.npy b/history_trace_back_v2.npy
similarity index 100%
rename from DRED_codegen/history_trace_back_v2.npy
rename to history_trace_back_v2.npy
diff --git a/DRED_codegen/hls_proj/.apc/autopilot.apfmapping b/hls_proj/.apc/autopilot.apfmapping
similarity index 100%
rename from DRED_codegen/hls_proj/.apc/autopilot.apfmapping
rename to hls_proj/.apc/autopilot.apfmapping
diff --git a/DRED_codegen/hls_proj/.cproject b/hls_proj/.cproject
similarity index 100%
rename from DRED_codegen/hls_proj/.cproject
rename to hls_proj/.cproject
diff --git a/DRED_codegen/hls_proj/.project b/hls_proj/.project
similarity index 100%
rename from DRED_codegen/hls_proj/.project
rename to hls_proj/.project
diff --git a/DRED_codegen/hls_proj/.settings/hls_proj.Debug.launch b/hls_proj/.settings/hls_proj.Debug.launch
similarity index 100%
rename from DRED_codegen/hls_proj/.settings/hls_proj.Debug.launch
rename to hls_proj/.settings/hls_proj.Debug.launch
diff --git a/DRED_codegen/hls_proj/.settings/hls_proj.Release.launch b/hls_proj/.settings/hls_proj.Release.launch
similarity index 100%
rename from DRED_codegen/hls_proj/.settings/hls_proj.Release.launch
rename to hls_proj/.settings/hls_proj.Release.launch
diff --git a/DRED_codegen/hls_proj/.vivado_hls_log_all.xml b/hls_proj/.vivado_hls_log_all.xml
similarity index 100%
rename from DRED_codegen/hls_proj/.vivado_hls_log_all.xml
rename to hls_proj/.vivado_hls_log_all.xml
diff --git a/DRED_codegen/hls_proj/gened.cpp b/hls_proj/gened.cpp
similarity index 100%
rename from DRED_codegen/hls_proj/gened.cpp
rename to hls_proj/gened.cpp
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.export.ll b/hls_proj/solution1/.autopilot/db/a.export.ll
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.export.ll
rename to hls_proj/solution1/.autopilot/db/a.export.ll
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g b/hls_proj/solution1/.autopilot/db/a.g
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.g
rename to hls_proj/solution1/.autopilot/db/a.g
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.0 b/hls_proj/solution1/.autopilot/db/a.g.0
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.0
rename to hls_proj/solution1/.autopilot/db/a.g.0
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.0.bc b/hls_proj/solution1/.autopilot/db/a.g.0.bc
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.0.bc
rename to hls_proj/solution1/.autopilot/db/a.g.0.bc
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.1.bc b/hls_proj/solution1/.autopilot/db/a.g.1.bc
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.1.bc
rename to hls_proj/solution1/.autopilot/db/a.g.1.bc
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.1.ll b/hls_proj/solution1/.autopilot/db/a.g.1.ll
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.1.ll
rename to hls_proj/solution1/.autopilot/db/a.g.1.ll
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.2.bc b/hls_proj/solution1/.autopilot/db/a.g.2.bc
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.2.bc
rename to hls_proj/solution1/.autopilot/db/a.g.2.bc
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.2.prechk.bc b/hls_proj/solution1/.autopilot/db/a.g.2.prechk.bc
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.2.prechk.bc
rename to hls_proj/solution1/.autopilot/db/a.g.2.prechk.bc
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.bc b/hls_proj/solution1/.autopilot/db/a.g.bc
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.bc
rename to hls_proj/solution1/.autopilot/db/a.g.bc
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.ll b/hls_proj/solution1/.autopilot/db/a.g.ll
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.g.ll
rename to hls_proj/solution1/.autopilot/db/a.g.ll
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o b/hls_proj/solution1/.autopilot/db/a.o
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.o
rename to hls_proj/solution1/.autopilot/db/a.o
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.1.bc b/hls_proj/solution1/.autopilot/db/a.o.1.bc
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.1.bc
rename to hls_proj/solution1/.autopilot/db/a.o.1.bc
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.1.ll b/hls_proj/solution1/.autopilot/db/a.o.1.ll
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.1.ll
rename to hls_proj/solution1/.autopilot/db/a.o.1.ll
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.1.tmp.bc b/hls_proj/solution1/.autopilot/db/a.o.1.tmp.bc
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.1.tmp.bc
rename to hls_proj/solution1/.autopilot/db/a.o.1.tmp.bc
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.2.bc b/hls_proj/solution1/.autopilot/db/a.o.2.bc
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.2.bc
rename to hls_proj/solution1/.autopilot/db/a.o.2.bc
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.2.ll b/hls_proj/solution1/.autopilot/db/a.o.2.ll
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.2.ll
rename to hls_proj/solution1/.autopilot/db/a.o.2.ll
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.3.bc b/hls_proj/solution1/.autopilot/db/a.o.3.bc
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.3.bc
rename to hls_proj/solution1/.autopilot/db/a.o.3.bc
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.3.ll b/hls_proj/solution1/.autopilot/db/a.o.3.ll
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.3.ll
rename to hls_proj/solution1/.autopilot/db/a.o.3.ll
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.bc b/hls_proj/solution1/.autopilot/db/a.o.bc
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.bc
rename to hls_proj/solution1/.autopilot/db/a.o.bc
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.ll b/hls_proj/solution1/.autopilot/db/a.o.ll
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.o.ll
rename to hls_proj/solution1/.autopilot/db/a.o.ll
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.pp.0.bc b/hls_proj/solution1/.autopilot/db/a.pp.0.bc
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.pp.0.bc
rename to hls_proj/solution1/.autopilot/db/a.pp.0.bc
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/a.pp.bc b/hls_proj/solution1/.autopilot/db/a.pp.bc
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/a.pp.bc
rename to hls_proj/solution1/.autopilot/db/a.pp.bc
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/autopilot.flow.log b/hls_proj/solution1/.autopilot/db/autopilot.flow.log
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/autopilot.flow.log
rename to hls_proj/solution1/.autopilot/db/autopilot.flow.log
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/autopilot.rtl.models.txt b/hls_proj/solution1/.autopilot/db/autopilot.rtl.models.txt
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/autopilot.rtl.models.txt
rename to hls_proj/solution1/.autopilot/db/autopilot.rtl.models.txt
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/bugpoint.sh b/hls_proj/solution1/.autopilot/db/bugpoint.sh
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/bugpoint.sh
rename to hls_proj/solution1/.autopilot/db/bugpoint.sh
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.bc b/hls_proj/solution1/.autopilot/db/gened.bc
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/gened.bc
rename to hls_proj/solution1/.autopilot/db/gened.bc
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.g.bc b/hls_proj/solution1/.autopilot/db/gened.g.bc
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/gened.g.bc
rename to hls_proj/solution1/.autopilot/db/gened.g.bc
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pp.0.cpp b/hls_proj/solution1/.autopilot/db/gened.pp.0.cpp
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pp.0.cpp
rename to hls_proj/solution1/.autopilot/db/gened.pp.0.cpp
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pp.0.cpp.ap-cdt.cpp b/hls_proj/solution1/.autopilot/db/gened.pp.0.cpp.ap-cdt.cpp
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pp.0.cpp.ap-cdt.cpp
rename to hls_proj/solution1/.autopilot/db/gened.pp.0.cpp.ap-cdt.cpp
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pp.0.cpp.ap-line.cpp b/hls_proj/solution1/.autopilot/db/gened.pp.0.cpp.ap-line.cpp
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pp.0.cpp.ap-line.cpp
rename to hls_proj/solution1/.autopilot/db/gened.pp.0.cpp.ap-line.cpp
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pp.0.cpp.ap-line.cpp.CXX b/hls_proj/solution1/.autopilot/db/gened.pp.0.cpp.ap-line.cpp.CXX
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pp.0.cpp.ap-line.cpp.CXX
rename to hls_proj/solution1/.autopilot/db/gened.pp.0.cpp.ap-line.cpp.CXX
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pp.00.o b/hls_proj/solution1/.autopilot/db/gened.pp.00.o
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pp.00.o
rename to hls_proj/solution1/.autopilot/db/gened.pp.00.o
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pragma.0.cpp b/hls_proj/solution1/.autopilot/db/gened.pragma.0.cpp
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pragma.0.cpp
rename to hls_proj/solution1/.autopilot/db/gened.pragma.0.cpp
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pragma.0.cpp.ap-line.CXX b/hls_proj/solution1/.autopilot/db/gened.pragma.0.cpp.ap-line.CXX
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pragma.0.cpp.ap-line.CXX
rename to hls_proj/solution1/.autopilot/db/gened.pragma.0.cpp.ap-line.CXX
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pragma.1.cpp b/hls_proj/solution1/.autopilot/db/gened.pragma.1.cpp
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pragma.1.cpp
rename to hls_proj/solution1/.autopilot/db/gened.pragma.1.cpp
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pragma.2.cpp b/hls_proj/solution1/.autopilot/db/gened.pragma.2.cpp
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/gened.pragma.2.cpp
rename to hls_proj/solution1/.autopilot/db/gened.pragma.2.cpp
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/global.setting.tcl b/hls_proj/solution1/.autopilot/db/global.setting.tcl
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/global.setting.tcl
rename to hls_proj/solution1/.autopilot/db/global.setting.tcl
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/hls_design_meta.cpp b/hls_proj/solution1/.autopilot/db/hls_design_meta.cpp
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/hls_design_meta.cpp
rename to hls_proj/solution1/.autopilot/db/hls_design_meta.cpp
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/hls_design_meta.h b/hls_proj/solution1/.autopilot/db/hls_design_meta.h
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/hls_design_meta.h
rename to hls_proj/solution1/.autopilot/db/hls_design_meta.h
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/hls_design_meta.tcl b/hls_proj/solution1/.autopilot/db/hls_design_meta.tcl
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/hls_design_meta.tcl
rename to hls_proj/solution1/.autopilot/db/hls_design_meta.tcl
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/pragma.0.directive b/hls_proj/solution1/.autopilot/db/pragma.0.directive
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/pragma.0.directive
rename to hls_proj/solution1/.autopilot/db/pragma.0.directive
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/pragma.1.directive b/hls_proj/solution1/.autopilot/db/pragma.1.directive
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/pragma.1.directive
rename to hls_proj/solution1/.autopilot/db/pragma.1.directive
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/pragma.status.tcl b/hls_proj/solution1/.autopilot/db/pragma.status.tcl
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/pragma.status.tcl
rename to hls_proj/solution1/.autopilot/db/pragma.status.tcl
diff --git a/DRED_codegen/hls_proj/solution1/.autopilot/db/ve_warning.tcl b/hls_proj/solution1/.autopilot/db/ve_warning.tcl
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/.autopilot/db/ve_warning.tcl
rename to hls_proj/solution1/.autopilot/db/ve_warning.tcl
diff --git a/DRED_codegen/hls_proj/solution1/solution1.aps b/hls_proj/solution1/solution1.aps
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/solution1.aps
rename to hls_proj/solution1/solution1.aps
diff --git a/DRED_codegen/hls_proj/solution1/solution1.directive b/hls_proj/solution1/solution1.directive
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/solution1.directive
rename to hls_proj/solution1/solution1.directive
diff --git a/DRED_codegen/hls_proj/solution1/solution1.log b/hls_proj/solution1/solution1.log
similarity index 100%
rename from DRED_codegen/hls_proj/solution1/solution1.log
rename to hls_proj/solution1/solution1.log
diff --git a/DRED_codegen/hls_proj/vivado_hls.app b/hls_proj/vivado_hls.app
similarity index 100%
rename from DRED_codegen/hls_proj/vivado_hls.app
rename to hls_proj/vivado_hls.app
diff --git a/DRED_codegen/inference.py b/inference.py
similarity index 100%
rename from DRED_codegen/inference.py
rename to inference.py
diff --git a/DRED_codegen/inference_escape.py b/inference_escape.py
similarity index 100%
rename from DRED_codegen/inference_escape.py
rename to inference_escape.py
diff --git a/nmtdecoder/ce.cc b/nmtdecoder/ce.cc
deleted file mode 100644
index 66bd146316e4a96325aed845c63f8425011f19db..0000000000000000000000000000000000000000
--- a/nmtdecoder/ce.cc
+++ /dev/null
@@ -1,3483 +0,0 @@
-#include "ce.h"
-float compute_engine8(
-float a1, float b1,
-float a2, float b2,
-float a3, float b3,
-float a4, float b4,
-float a5, float b5,
-float a6, float b6,
-float a7, float b7,
-float a8, float b8){
-#pragma HLS PIPELINE
- float m1, m2, m3, m4,
- m5, m6, m7, m8;
-
- float add1, add2, add3, add4,
- add5, add6;
- m1 = a1*b1;
- m2 = a2*b2;
- m3 = a3*b3;
- m4 = a4*b4;
- m5 = a5*b5;
- m6 = a6*b6;
- m7 = a7*b7;
- m8 = a8*b8;
- add1 = m1+m2;
- add2 = m3+m4;
- add3 = m5+m6;
- add4 = m7+m8;
- add5 = add4+add3;
- add6 = add2+add1;
- return add5+add6;
-}
-float compute_engine16(
-float a1, float b1,
-float a2, float b2,
-float a3, float b3,
-float a4, float b4,
-float a5, float b5,
-float a6, float b6,
-float a7, float b7,
-float a8, float b8,
-float a9, float b9,
-float a10, float b10,
-float a11, float b11,
-float a12, float b12,
-float a13, float b13,
-float a14, float b14,
-float a15, float b15,
-float a16, float b16){
-#pragma HLS PIPELINE
- float m1, m2, m3, m4,
- m5, m6, m7, m8, m9,
- m10, m11, m12, m13, m14,
- m15, m16;
-
- float add1, add2, add3, add4,
- add5, add6, add7, add8, add9,
- add10, add11, add12, add13, add14;
- m1 = a1*b1;
- m2 = a2*b2;
- m3 = a3*b3;
- m4 = a4*b4;
- m5 = a5*b5;
- m6 = a6*b6;
- m7 = a7*b7;
- m8 = a8*b8;
- m9 = a9*b9;
- m10 = a10*b10;
- m11 = a11*b11;
- m12 = a12*b12;
- m13 = a13*b13;
- m14 = a14*b14;
- m15 = a15*b15;
- m16 = a16*b16;
- add1 = m1+m2;
- add2 = m3+m4;
- add3 = m5+m6;
- add4 = m7+m8;
- add5 = m9+m10;
- add6 = m11+m12;
- add7 = m13+m14;
- add8 = m15+m16;
- add9 = add8+add7;
- add10 = add6+add5;
- add11 = add4+add3;
- add12 = add2+add1;
- add13 = add12+add11;
- add14 = add10+add9;
- return add13+add14;
-}
-float compute_engine32(
-float a1, float b1,
-float a2, float b2,
-float a3, float b3,
-float a4, float b4,
-float a5, float b5,
-float a6, float b6,
-float a7, float b7,
-float a8, float b8,
-float a9, float b9,
-float a10, float b10,
-float a11, float b11,
-float a12, float b12,
-float a13, float b13,
-float a14, float b14,
-float a15, float b15,
-float a16, float b16,
-float a17, float b17,
-float a18, float b18,
-float a19, float b19,
-float a20, float b20,
-float a21, float b21,
-float a22, float b22,
-float a23, float b23,
-float a24, float b24,
-float a25, float b25,
-float a26, float b26,
-float a27, float b27,
-float a28, float b28,
-float a29, float b29,
-float a30, float b30,
-float a31, float b31,
-float a32, float b32){
-#pragma HLS PIPELINE
- float m1, m2, m3, m4,
- m5, m6, m7, m8, m9,
- m10, m11, m12, m13, m14,
- m15, m16, m17, m18, m19,
- m20, m21, m22, m23, m24,
- m25, m26, m27, m28, m29,
- m30, m31, m32;
-
- float add1, add2, add3, add4,
- add5, add6, add7, add8, add9,
- add10, add11, add12, add13, add14,
- add15, add16, add17, add18, add19,
- add20, add21, add22, add23, add24,
- add25, add26, add27, add28, add29,
- add30;
- m1 = a1*b1;
- m2 = a2*b2;
- m3 = a3*b3;
- m4 = a4*b4;
- m5 = a5*b5;
- m6 = a6*b6;
- m7 = a7*b7;
- m8 = a8*b8;
- m9 = a9*b9;
- m10 = a10*b10;
- m11 = a11*b11;
- m12 = a12*b12;
- m13 = a13*b13;
- m14 = a14*b14;
- m15 = a15*b15;
- m16 = a16*b16;
- m17 = a17*b17;
- m18 = a18*b18;
- m19 = a19*b19;
- m20 = a20*b20;
- m21 = a21*b21;
- m22 = a22*b22;
- m23 = a23*b23;
- m24 = a24*b24;
- m25 = a25*b25;
- m26 = a26*b26;
- m27 = a27*b27;
- m28 = a28*b28;
- m29 = a29*b29;
- m30 = a30*b30;
- m31 = a31*b31;
- m32 = a32*b32;
- add1 = m1+m2;
- add2 = m3+m4;
- add3 = m5+m6;
- add4 = m7+m8;
- add5 = m9+m10;
- add6 = m11+m12;
- add7 = m13+m14;
- add8 = m15+m16;
- add9 = m17+m18;
- add10 = m19+m20;
- add11 = m21+m22;
- add12 = m23+m24;
- add13 = m25+m26;
- add14 = m27+m28;
- add15 = m29+m30;
- add16 = m31+m32;
- add17 = add16+add15;
- add18 = add14+add13;
- add19 = add12+add11;
- add20 = add10+add9;
- add21 = add8+add7;
- add22 = add6+add5;
- add23 = add4+add3;
- add24 = add2+add1;
- add25 = add24+add23;
- add26 = add22+add21;
- add27 = add20+add19;
- add28 = add18+add17;
- add29 = add28+add27;
- add30 = add26+add25;
- return add29+add30;
-}
-float compute_engine64(
-float a1, float b1,
-float a2, float b2,
-float a3, float b3,
-float a4, float b4,
-float a5, float b5,
-float a6, float b6,
-float a7, float b7,
-float a8, float b8,
-float a9, float b9,
-float a10, float b10,
-float a11, float b11,
-float a12, float b12,
-float a13, float b13,
-float a14, float b14,
-float a15, float b15,
-float a16, float b16,
-float a17, float b17,
-float a18, float b18,
-float a19, float b19,
-float a20, float b20,
-float a21, float b21,
-float a22, float b22,
-float a23, float b23,
-float a24, float b24,
-float a25, float b25,
-float a26, float b26,
-float a27, float b27,
-float a28, float b28,
-float a29, float b29,
-float a30, float b30,
-float a31, float b31,
-float a32, float b32,
-float a33, float b33,
-float a34, float b34,
-float a35, float b35,
-float a36, float b36,
-float a37, float b37,
-float a38, float b38,
-float a39, float b39,
-float a40, float b40,
-float a41, float b41,
-float a42, float b42,
-float a43, float b43,
-float a44, float b44,
-float a45, float b45,
-float a46, float b46,
-float a47, float b47,
-float a48, float b48,
-float a49, float b49,
-float a50, float b50,
-float a51, float b51,
-float a52, float b52,
-float a53, float b53,
-float a54, float b54,
-float a55, float b55,
-float a56, float b56,
-float a57, float b57,
-float a58, float b58,
-float a59, float b59,
-float a60, float b60,
-float a61, float b61,
-float a62, float b62,
-float a63, float b63,
-float a64, float b64){
-#pragma HLS inline off
-#pragma HLS PIPELINE
- float m1, m2, m3, m4,
- m5, m6, m7, m8, m9,
- m10, m11, m12, m13, m14,
- m15, m16, m17, m18, m19,
- m20, m21, m22, m23, m24,
- m25, m26, m27, m28, m29,
- m30, m31, m32, m33, m34,
- m35, m36, m37, m38, m39,
- m40, m41, m42, m43, m44,
- m45, m46, m47, m48, m49,
- m50, m51, m52, m53, m54,
- m55, m56, m57, m58, m59,
- m60, m61, m62, m63, m64;
-
- float add1, add2, add3, add4,
- add5, add6, add7, add8, add9,
- add10, add11, add12, add13, add14,
- add15, add16, add17, add18, add19,
- add20, add21, add22, add23, add24,
- add25, add26, add27, add28, add29,
- add30, add31, add32, add33, add34,
- add35, add36, add37, add38, add39,
- add40, add41, add42, add43, add44,
- add45, add46, add47, add48, add49,
- add50, add51, add52, add53, add54,
- add55, add56, add57, add58, add59,
- add60, add61, add62;
- m1 = a1*b1;
- m2 = a2*b2;
- m3 = a3*b3;
- m4 = a4*b4;
- m5 = a5*b5;
- m6 = a6*b6;
- m7 = a7*b7;
- m8 = a8*b8;
- m9 = a9*b9;
- m10 = a10*b10;
- m11 = a11*b11;
- m12 = a12*b12;
- m13 = a13*b13;
- m14 = a14*b14;
- m15 = a15*b15;
- m16 = a16*b16;
- m17 = a17*b17;
- m18 = a18*b18;
- m19 = a19*b19;
- m20 = a20*b20;
- m21 = a21*b21;
- m22 = a22*b22;
- m23 = a23*b23;
- m24 = a24*b24;
- m25 = a25*b25;
- m26 = a26*b26;
- m27 = a27*b27;
- m28 = a28*b28;
- m29 = a29*b29;
- m30 = a30*b30;
- m31 = a31*b31;
- m32 = a32*b32;
- m33 = a33*b33;
- m34 = a34*b34;
- m35 = a35*b35;
- m36 = a36*b36;
- m37 = a37*b37;
- m38 = a38*b38;
- m39 = a39*b39;
- m40 = a40*b40;
- m41 = a41*b41;
- m42 = a42*b42;
- m43 = a43*b43;
- m44 = a44*b44;
- m45 = a45*b45;
- m46 = a46*b46;
- m47 = a47*b47;
- m48 = a48*b48;
- m49 = a49*b49;
- m50 = a50*b50;
- m51 = a51*b51;
- m52 = a52*b52;
- m53 = a53*b53;
- m54 = a54*b54;
- m55 = a55*b55;
- m56 = a56*b56;
- m57 = a57*b57;
- m58 = a58*b58;
- m59 = a59*b59;
- m60 = a60*b60;
- m61 = a61*b61;
- m62 = a62*b62;
- m63 = a63*b63;
- m64 = a64*b64;
- add1 = m1+m2;
- add2 = m3+m4;
- add3 = m5+m6;
- add4 = m7+m8;
- add5 = m9+m10;
- add6 = m11+m12;
- add7 = m13+m14;
- add8 = m15+m16;
- add9 = m17+m18;
- add10 = m19+m20;
- add11 = m21+m22;
- add12 = m23+m24;
- add13 = m25+m26;
- add14 = m27+m28;
- add15 = m29+m30;
- add16 = m31+m32;
- add17 = m33+m34;
- add18 = m35+m36;
- add19 = m37+m38;
- add20 = m39+m40;
- add21 = m41+m42;
- add22 = m43+m44;
- add23 = m45+m46;
- add24 = m47+m48;
- add25 = m49+m50;
- add26 = m51+m52;
- add27 = m53+m54;
- add28 = m55+m56;
- add29 = m57+m58;
- add30 = m59+m60;
- add31 = m61+m62;
- add32 = m63+m64;
- add33 = add32+add31;
- add34 = add30+add29;
- add35 = add28+add27;
- add36 = add26+add25;
- add37 = add24+add23;
- add38 = add22+add21;
- add39 = add20+add19;
- add40 = add18+add17;
- add41 = add16+add15;
- add42 = add14+add13;
- add43 = add12+add11;
- add44 = add10+add9;
- add45 = add8+add7;
- add46 = add6+add5;
- add47 = add4+add3;
- add48 = add2+add1;
- add49 = add48+add47;
- add50 = add46+add45;
- add51 = add44+add43;
- add52 = add42+add41;
- add53 = add40+add39;
- add54 = add38+add37;
- add55 = add36+add35;
- add56 = add34+add33;
- add57 = add56+add55;
- add58 = add54+add53;
- add59 = add52+add51;
- add60 = add50+add49;
- add61 = add60+add59;
- add62 = add58+add57;
- return add61+add62;
-}
-float compute_engine128(
-float a1, float b1,
-float a2, float b2,
-float a3, float b3,
-float a4, float b4,
-float a5, float b5,
-float a6, float b6,
-float a7, float b7,
-float a8, float b8,
-float a9, float b9,
-float a10, float b10,
-float a11, float b11,
-float a12, float b12,
-float a13, float b13,
-float a14, float b14,
-float a15, float b15,
-float a16, float b16,
-float a17, float b17,
-float a18, float b18,
-float a19, float b19,
-float a20, float b20,
-float a21, float b21,
-float a22, float b22,
-float a23, float b23,
-float a24, float b24,
-float a25, float b25,
-float a26, float b26,
-float a27, float b27,
-float a28, float b28,
-float a29, float b29,
-float a30, float b30,
-float a31, float b31,
-float a32, float b32,
-float a33, float b33,
-float a34, float b34,
-float a35, float b35,
-float a36, float b36,
-float a37, float b37,
-float a38, float b38,
-float a39, float b39,
-float a40, float b40,
-float a41, float b41,
-float a42, float b42,
-float a43, float b43,
-float a44, float b44,
-float a45, float b45,
-float a46, float b46,
-float a47, float b47,
-float a48, float b48,
-float a49, float b49,
-float a50, float b50,
-float a51, float b51,
-float a52, float b52,
-float a53, float b53,
-float a54, float b54,
-float a55, float b55,
-float a56, float b56,
-float a57, float b57,
-float a58, float b58,
-float a59, float b59,
-float a60, float b60,
-float a61, float b61,
-float a62, float b62,
-float a63, float b63,
-float a64, float b64,
-float a65, float b65,
-float a66, float b66,
-float a67, float b67,
-float a68, float b68,
-float a69, float b69,
-float a70, float b70,
-float a71, float b71,
-float a72, float b72,
-float a73, float b73,
-float a74, float b74,
-float a75, float b75,
-float a76, float b76,
-float a77, float b77,
-float a78, float b78,
-float a79, float b79,
-float a80, float b80,
-float a81, float b81,
-float a82, float b82,
-float a83, float b83,
-float a84, float b84,
-float a85, float b85,
-float a86, float b86,
-float a87, float b87,
-float a88, float b88,
-float a89, float b89,
-float a90, float b90,
-float a91, float b91,
-float a92, float b92,
-float a93, float b93,
-float a94, float b94,
-float a95, float b95,
-float a96, float b96,
-float a97, float b97,
-float a98, float b98,
-float a99, float b99,
-float a100, float b100,
-float a101, float b101,
-float a102, float b102,
-float a103, float b103,
-float a104, float b104,
-float a105, float b105,
-float a106, float b106,
-float a107, float b107,
-float a108, float b108,
-float a109, float b109,
-float a110, float b110,
-float a111, float b111,
-float a112, float b112,
-float a113, float b113,
-float a114, float b114,
-float a115, float b115,
-float a116, float b116,
-float a117, float b117,
-float a118, float b118,
-float a119, float b119,
-float a120, float b120,
-float a121, float b121,
-float a122, float b122,
-float a123, float b123,
-float a124, float b124,
-float a125, float b125,
-float a126, float b126,
-float a127, float b127,
-float a128, float b128){
-#pragma HLS PIPELINE
- float m1, m2, m3, m4,
- m5, m6, m7, m8, m9,
- m10, m11, m12, m13, m14,
- m15, m16, m17, m18, m19,
- m20, m21, m22, m23, m24,
- m25, m26, m27, m28, m29,
- m30, m31, m32, m33, m34,
- m35, m36, m37, m38, m39,
- m40, m41, m42, m43, m44,
- m45, m46, m47, m48, m49,
- m50, m51, m52, m53, m54,
- m55, m56, m57, m58, m59,
- m60, m61, m62, m63, m64,
- m65, m66, m67, m68, m69,
- m70, m71, m72, m73, m74,
- m75, m76, m77, m78, m79,
- m80, m81, m82, m83, m84,
- m85, m86, m87, m88, m89,
- m90, m91, m92, m93, m94,
- m95, m96, m97, m98, m99,
- m100, m101, m102, m103, m104,
- m105, m106, m107, m108, m109,
- m110, m111, m112, m113, m114,
- m115, m116, m117, m118, m119,
- m120, m121, m122, m123, m124,
- m125, m126, m127, m128;
-
- float add1, add2, add3, add4,
- add5, add6, add7, add8, add9,
- add10, add11, add12, add13, add14,
- add15, add16, add17, add18, add19,
- add20, add21, add22, add23, add24,
- add25, add26, add27, add28, add29,
- add30, add31, add32, add33, add34,
- add35, add36, add37, add38, add39,
- add40, add41, add42, add43, add44,
- add45, add46, add47, add48, add49,
- add50, add51, add52, add53, add54,
- add55, add56, add57, add58, add59,
- add60, add61, add62, add63, add64,
- add65, add66, add67, add68, add69,
- add70, add71, add72, add73, add74,
- add75, add76, add77, add78, add79,
- add80, add81, add82, add83, add84,
- add85, add86, add87, add88, add89,
- add90, add91, add92, add93, add94,
- add95, add96, add97, add98, add99,
- add100, add101, add102, add103, add104,
- add105, add106, add107, add108, add109,
- add110, add111, add112, add113, add114,
- add115, add116, add117, add118, add119,
- add120, add121, add122, add123, add124,
- add125, add126;
- m1 = a1*b1;
- m2 = a2*b2;
- m3 = a3*b3;
- m4 = a4*b4;
- m5 = a5*b5;
- m6 = a6*b6;
- m7 = a7*b7;
- m8 = a8*b8;
- m9 = a9*b9;
- m10 = a10*b10;
- m11 = a11*b11;
- m12 = a12*b12;
- m13 = a13*b13;
- m14 = a14*b14;
- m15 = a15*b15;
- m16 = a16*b16;
- m17 = a17*b17;
- m18 = a18*b18;
- m19 = a19*b19;
- m20 = a20*b20;
- m21 = a21*b21;
- m22 = a22*b22;
- m23 = a23*b23;
- m24 = a24*b24;
- m25 = a25*b25;
- m26 = a26*b26;
- m27 = a27*b27;
- m28 = a28*b28;
- m29 = a29*b29;
- m30 = a30*b30;
- m31 = a31*b31;
- m32 = a32*b32;
- m33 = a33*b33;
- m34 = a34*b34;
- m35 = a35*b35;
- m36 = a36*b36;
- m37 = a37*b37;
- m38 = a38*b38;
- m39 = a39*b39;
- m40 = a40*b40;
- m41 = a41*b41;
- m42 = a42*b42;
- m43 = a43*b43;
- m44 = a44*b44;
- m45 = a45*b45;
- m46 = a46*b46;
- m47 = a47*b47;
- m48 = a48*b48;
- m49 = a49*b49;
- m50 = a50*b50;
- m51 = a51*b51;
- m52 = a52*b52;
- m53 = a53*b53;
- m54 = a54*b54;
- m55 = a55*b55;
- m56 = a56*b56;
- m57 = a57*b57;
- m58 = a58*b58;
- m59 = a59*b59;
- m60 = a60*b60;
- m61 = a61*b61;
- m62 = a62*b62;
- m63 = a63*b63;
- m64 = a64*b64;
- m65 = a65*b65;
- m66 = a66*b66;
- m67 = a67*b67;
- m68 = a68*b68;
- m69 = a69*b69;
- m70 = a70*b70;
- m71 = a71*b71;
- m72 = a72*b72;
- m73 = a73*b73;
- m74 = a74*b74;
- m75 = a75*b75;
- m76 = a76*b76;
- m77 = a77*b77;
- m78 = a78*b78;
- m79 = a79*b79;
- m80 = a80*b80;
- m81 = a81*b81;
- m82 = a82*b82;
- m83 = a83*b83;
- m84 = a84*b84;
- m85 = a85*b85;
- m86 = a86*b86;
- m87 = a87*b87;
- m88 = a88*b88;
- m89 = a89*b89;
- m90 = a90*b90;
- m91 = a91*b91;
- m92 = a92*b92;
- m93 = a93*b93;
- m94 = a94*b94;
- m95 = a95*b95;
- m96 = a96*b96;
- m97 = a97*b97;
- m98 = a98*b98;
- m99 = a99*b99;
- m100 = a100*b100;
- m101 = a101*b101;
- m102 = a102*b102;
- m103 = a103*b103;
- m104 = a104*b104;
- m105 = a105*b105;
- m106 = a106*b106;
- m107 = a107*b107;
- m108 = a108*b108;
- m109 = a109*b109;
- m110 = a110*b110;
- m111 = a111*b111;
- m112 = a112*b112;
- m113 = a113*b113;
- m114 = a114*b114;
- m115 = a115*b115;
- m116 = a116*b116;
- m117 = a117*b117;
- m118 = a118*b118;
- m119 = a119*b119;
- m120 = a120*b120;
- m121 = a121*b121;
- m122 = a122*b122;
- m123 = a123*b123;
- m124 = a124*b124;
- m125 = a125*b125;
- m126 = a126*b126;
- m127 = a127*b127;
- m128 = a128*b128;
- add1 = m1+m2;
- add2 = m3+m4;
- add3 = m5+m6;
- add4 = m7+m8;
- add5 = m9+m10;
- add6 = m11+m12;
- add7 = m13+m14;
- add8 = m15+m16;
- add9 = m17+m18;
- add10 = m19+m20;
- add11 = m21+m22;
- add12 = m23+m24;
- add13 = m25+m26;
- add14 = m27+m28;
- add15 = m29+m30;
- add16 = m31+m32;
- add17 = m33+m34;
- add18 = m35+m36;
- add19 = m37+m38;
- add20 = m39+m40;
- add21 = m41+m42;
- add22 = m43+m44;
- add23 = m45+m46;
- add24 = m47+m48;
- add25 = m49+m50;
- add26 = m51+m52;
- add27 = m53+m54;
- add28 = m55+m56;
- add29 = m57+m58;
- add30 = m59+m60;
- add31 = m61+m62;
- add32 = m63+m64;
- add33 = m65+m66;
- add34 = m67+m68;
- add35 = m69+m70;
- add36 = m71+m72;
- add37 = m73+m74;
- add38 = m75+m76;
- add39 = m77+m78;
- add40 = m79+m80;
- add41 = m81+m82;
- add42 = m83+m84;
- add43 = m85+m86;
- add44 = m87+m88;
- add45 = m89+m90;
- add46 = m91+m92;
- add47 = m93+m94;
- add48 = m95+m96;
- add49 = m97+m98;
- add50 = m99+m100;
- add51 = m101+m102;
- add52 = m103+m104;
- add53 = m105+m106;
- add54 = m107+m108;
- add55 = m109+m110;
- add56 = m111+m112;
- add57 = m113+m114;
- add58 = m115+m116;
- add59 = m117+m118;
- add60 = m119+m120;
- add61 = m121+m122;
- add62 = m123+m124;
- add63 = m125+m126;
- add64 = m127+m128;
- add65 = add64+add63;
- add66 = add62+add61;
- add67 = add60+add59;
- add68 = add58+add57;
- add69 = add56+add55;
- add70 = add54+add53;
- add71 = add52+add51;
- add72 = add50+add49;
- add73 = add48+add47;
- add74 = add46+add45;
- add75 = add44+add43;
- add76 = add42+add41;
- add77 = add40+add39;
- add78 = add38+add37;
- add79 = add36+add35;
- add80 = add34+add33;
- add81 = add32+add31;
- add82 = add30+add29;
- add83 = add28+add27;
- add84 = add26+add25;
- add85 = add24+add23;
- add86 = add22+add21;
- add87 = add20+add19;
- add88 = add18+add17;
- add89 = add16+add15;
- add90 = add14+add13;
- add91 = add12+add11;
- add92 = add10+add9;
- add93 = add8+add7;
- add94 = add6+add5;
- add95 = add4+add3;
- add96 = add2+add1;
- add97 = add96+add95;
- add98 = add94+add93;
- add99 = add92+add91;
- add100 = add90+add89;
- add101 = add88+add87;
- add102 = add86+add85;
- add103 = add84+add83;
- add104 = add82+add81;
- add105 = add80+add79;
- add106 = add78+add77;
- add107 = add76+add75;
- add108 = add74+add73;
- add109 = add72+add71;
- add110 = add70+add69;
- add111 = add68+add67;
- add112 = add66+add65;
- add113 = add112+add111;
- add114 = add110+add109;
- add115 = add108+add107;
- add116 = add106+add105;
- add117 = add104+add103;
- add118 = add102+add101;
- add119 = add100+add99;
- add120 = add98+add97;
- add121 = add120+add119;
- add122 = add118+add117;
- add123 = add116+add115;
- add124 = add114+add113;
- add125 = add124+add123;
- add126 = add122+add121;
- return add125+add126;
-}
-float compute_engine256(
-float a1, float b1,
-float a2, float b2,
-float a3, float b3,
-float a4, float b4,
-float a5, float b5,
-float a6, float b6,
-float a7, float b7,
-float a8, float b8,
-float a9, float b9,
-float a10, float b10,
-float a11, float b11,
-float a12, float b12,
-float a13, float b13,
-float a14, float b14,
-float a15, float b15,
-float a16, float b16,
-float a17, float b17,
-float a18, float b18,
-float a19, float b19,
-float a20, float b20,
-float a21, float b21,
-float a22, float b22,
-float a23, float b23,
-float a24, float b24,
-float a25, float b25,
-float a26, float b26,
-float a27, float b27,
-float a28, float b28,
-float a29, float b29,
-float a30, float b30,
-float a31, float b31,
-float a32, float b32,
-float a33, float b33,
-float a34, float b34,
-float a35, float b35,
-float a36, float b36,
-float a37, float b37,
-float a38, float b38,
-float a39, float b39,
-float a40, float b40,
-float a41, float b41,
-float a42, float b42,
-float a43, float b43,
-float a44, float b44,
-float a45, float b45,
-float a46, float b46,
-float a47, float b47,
-float a48, float b48,
-float a49, float b49,
-float a50, float b50,
-float a51, float b51,
-float a52, float b52,
-float a53, float b53,
-float a54, float b54,
-float a55, float b55,
-float a56, float b56,
-float a57, float b57,
-float a58, float b58,
-float a59, float b59,
-float a60, float b60,
-float a61, float b61,
-float a62, float b62,
-float a63, float b63,
-float a64, float b64,
-float a65, float b65,
-float a66, float b66,
-float a67, float b67,
-float a68, float b68,
-float a69, float b69,
-float a70, float b70,
-float a71, float b71,
-float a72, float b72,
-float a73, float b73,
-float a74, float b74,
-float a75, float b75,
-float a76, float b76,
-float a77, float b77,
-float a78, float b78,
-float a79, float b79,
-float a80, float b80,
-float a81, float b81,
-float a82, float b82,
-float a83, float b83,
-float a84, float b84,
-float a85, float b85,
-float a86, float b86,
-float a87, float b87,
-float a88, float b88,
-float a89, float b89,
-float a90, float b90,
-float a91, float b91,
-float a92, float b92,
-float a93, float b93,
-float a94, float b94,
-float a95, float b95,
-float a96, float b96,
-float a97, float b97,
-float a98, float b98,
-float a99, float b99,
-float a100, float b100,
-float a101, float b101,
-float a102, float b102,
-float a103, float b103,
-float a104, float b104,
-float a105, float b105,
-float a106, float b106,
-float a107, float b107,
-float a108, float b108,
-float a109, float b109,
-float a110, float b110,
-float a111, float b111,
-float a112, float b112,
-float a113, float b113,
-float a114, float b114,
-float a115, float b115,
-float a116, float b116,
-float a117, float b117,
-float a118, float b118,
-float a119, float b119,
-float a120, float b120,
-float a121, float b121,
-float a122, float b122,
-float a123, float b123,
-float a124, float b124,
-float a125, float b125,
-float a126, float b126,
-float a127, float b127,
-float a128, float b128,
-float a129, float b129,
-float a130, float b130,
-float a131, float b131,
-float a132, float b132,
-float a133, float b133,
-float a134, float b134,
-float a135, float b135,
-float a136, float b136,
-float a137, float b137,
-float a138, float b138,
-float a139, float b139,
-float a140, float b140,
-float a141, float b141,
-float a142, float b142,
-float a143, float b143,
-float a144, float b144,
-float a145, float b145,
-float a146, float b146,
-float a147, float b147,
-float a148, float b148,
-float a149, float b149,
-float a150, float b150,
-float a151, float b151,
-float a152, float b152,
-float a153, float b153,
-float a154, float b154,
-float a155, float b155,
-float a156, float b156,
-float a157, float b157,
-float a158, float b158,
-float a159, float b159,
-float a160, float b160,
-float a161, float b161,
-float a162, float b162,
-float a163, float b163,
-float a164, float b164,
-float a165, float b165,
-float a166, float b166,
-float a167, float b167,
-float a168, float b168,
-float a169, float b169,
-float a170, float b170,
-float a171, float b171,
-float a172, float b172,
-float a173, float b173,
-float a174, float b174,
-float a175, float b175,
-float a176, float b176,
-float a177, float b177,
-float a178, float b178,
-float a179, float b179,
-float a180, float b180,
-float a181, float b181,
-float a182, float b182,
-float a183, float b183,
-float a184, float b184,
-float a185, float b185,
-float a186, float b186,
-float a187, float b187,
-float a188, float b188,
-float a189, float b189,
-float a190, float b190,
-float a191, float b191,
-float a192, float b192,
-float a193, float b193,
-float a194, float b194,
-float a195, float b195,
-float a196, float b196,
-float a197, float b197,
-float a198, float b198,
-float a199, float b199,
-float a200, float b200,
-float a201, float b201,
-float a202, float b202,
-float a203, float b203,
-float a204, float b204,
-float a205, float b205,
-float a206, float b206,
-float a207, float b207,
-float a208, float b208,
-float a209, float b209,
-float a210, float b210,
-float a211, float b211,
-float a212, float b212,
-float a213, float b213,
-float a214, float b214,
-float a215, float b215,
-float a216, float b216,
-float a217, float b217,
-float a218, float b218,
-float a219, float b219,
-float a220, float b220,
-float a221, float b221,
-float a222, float b222,
-float a223, float b223,
-float a224, float b224,
-float a225, float b225,
-float a226, float b226,
-float a227, float b227,
-float a228, float b228,
-float a229, float b229,
-float a230, float b230,
-float a231, float b231,
-float a232, float b232,
-float a233, float b233,
-float a234, float b234,
-float a235, float b235,
-float a236, float b236,
-float a237, float b237,
-float a238, float b238,
-float a239, float b239,
-float a240, float b240,
-float a241, float b241,
-float a242, float b242,
-float a243, float b243,
-float a244, float b244,
-float a245, float b245,
-float a246, float b246,
-float a247, float b247,
-float a248, float b248,
-float a249, float b249,
-float a250, float b250,
-float a251, float b251,
-float a252, float b252,
-float a253, float b253,
-float a254, float b254,
-float a255, float b255,
-float a256, float b256){
-#pragma HLS PIPELINE
- float m1, m2, m3, m4,
- m5, m6, m7, m8, m9,
- m10, m11, m12, m13, m14,
- m15, m16, m17, m18, m19,
- m20, m21, m22, m23, m24,
- m25, m26, m27, m28, m29,
- m30, m31, m32, m33, m34,
- m35, m36, m37, m38, m39,
- m40, m41, m42, m43, m44,
- m45, m46, m47, m48, m49,
- m50, m51, m52, m53, m54,
- m55, m56, m57, m58, m59,
- m60, m61, m62, m63, m64,
- m65, m66, m67, m68, m69,
- m70, m71, m72, m73, m74,
- m75, m76, m77, m78, m79,
- m80, m81, m82, m83, m84,
- m85, m86, m87, m88, m89,
- m90, m91, m92, m93, m94,
- m95, m96, m97, m98, m99,
- m100, m101, m102, m103, m104,
- m105, m106, m107, m108, m109,
- m110, m111, m112, m113, m114,
- m115, m116, m117, m118, m119,
- m120, m121, m122, m123, m124,
- m125, m126, m127, m128, m129,
- m130, m131, m132, m133, m134,
- m135, m136, m137, m138, m139,
- m140, m141, m142, m143, m144,
- m145, m146, m147, m148, m149,
- m150, m151, m152, m153, m154,
- m155, m156, m157, m158, m159,
- m160, m161, m162, m163, m164,
- m165, m166, m167, m168, m169,
- m170, m171, m172, m173, m174,
- m175, m176, m177, m178, m179,
- m180, m181, m182, m183, m184,
- m185, m186, m187, m188, m189,
- m190, m191, m192, m193, m194,
- m195, m196, m197, m198, m199,
- m200, m201, m202, m203, m204,
- m205, m206, m207, m208, m209,
- m210, m211, m212, m213, m214,
- m215, m216, m217, m218, m219,
- m220, m221, m222, m223, m224,
- m225, m226, m227, m228, m229,
- m230, m231, m232, m233, m234,
- m235, m236, m237, m238, m239,
- m240, m241, m242, m243, m244,
- m245, m246, m247, m248, m249,
- m250, m251, m252, m253, m254,
- m255, m256;
-
- float add1, add2, add3, add4,
- add5, add6, add7, add8, add9,
- add10, add11, add12, add13, add14,
- add15, add16, add17, add18, add19,
- add20, add21, add22, add23, add24,
- add25, add26, add27, add28, add29,
- add30, add31, add32, add33, add34,
- add35, add36, add37, add38, add39,
- add40, add41, add42, add43, add44,
- add45, add46, add47, add48, add49,
- add50, add51, add52, add53, add54,
- add55, add56, add57, add58, add59,
- add60, add61, add62, add63, add64,
- add65, add66, add67, add68, add69,
- add70, add71, add72, add73, add74,
- add75, add76, add77, add78, add79,
- add80, add81, add82, add83, add84,
- add85, add86, add87, add88, add89,
- add90, add91, add92, add93, add94,
- add95, add96, add97, add98, add99,
- add100, add101, add102, add103, add104,
- add105, add106, add107, add108, add109,
- add110, add111, add112, add113, add114,
- add115, add116, add117, add118, add119,
- add120, add121, add122, add123, add124,
- add125, add126, add127, add128, add129,
- add130, add131, add132, add133, add134,
- add135, add136, add137, add138, add139,
- add140, add141, add142, add143, add144,
- add145, add146, add147, add148, add149,
- add150, add151, add152, add153, add154,
- add155, add156, add157, add158, add159,
- add160, add161, add162, add163, add164,
- add165, add166, add167, add168, add169,
- add170, add171, add172, add173, add174,
- add175, add176, add177, add178, add179,
- add180, add181, add182, add183, add184,
- add185, add186, add187, add188, add189,
- add190, add191, add192, add193, add194,
- add195, add196, add197, add198, add199,
- add200, add201, add202, add203, add204,
- add205, add206, add207, add208, add209,
- add210, add211, add212, add213, add214,
- add215, add216, add217, add218, add219,
- add220, add221, add222, add223, add224,
- add225, add226, add227, add228, add229,
- add230, add231, add232, add233, add234,
- add235, add236, add237, add238, add239,
- add240, add241, add242, add243, add244,
- add245, add246, add247, add248, add249,
- add250, add251, add252, add253, add254;
- m1 = a1*b1;
- m2 = a2*b2;
- m3 = a3*b3;
- m4 = a4*b4;
- m5 = a5*b5;
- m6 = a6*b6;
- m7 = a7*b7;
- m8 = a8*b8;
- m9 = a9*b9;
- m10 = a10*b10;
- m11 = a11*b11;
- m12 = a12*b12;
- m13 = a13*b13;
- m14 = a14*b14;
- m15 = a15*b15;
- m16 = a16*b16;
- m17 = a17*b17;
- m18 = a18*b18;
- m19 = a19*b19;
- m20 = a20*b20;
- m21 = a21*b21;
- m22 = a22*b22;
- m23 = a23*b23;
- m24 = a24*b24;
- m25 = a25*b25;
- m26 = a26*b26;
- m27 = a27*b27;
- m28 = a28*b28;
- m29 = a29*b29;
- m30 = a30*b30;
- m31 = a31*b31;
- m32 = a32*b32;
- m33 = a33*b33;
- m34 = a34*b34;
- m35 = a35*b35;
- m36 = a36*b36;
- m37 = a37*b37;
- m38 = a38*b38;
- m39 = a39*b39;
- m40 = a40*b40;
- m41 = a41*b41;
- m42 = a42*b42;
- m43 = a43*b43;
- m44 = a44*b44;
- m45 = a45*b45;
- m46 = a46*b46;
- m47 = a47*b47;
- m48 = a48*b48;
- m49 = a49*b49;
- m50 = a50*b50;
- m51 = a51*b51;
- m52 = a52*b52;
- m53 = a53*b53;
- m54 = a54*b54;
- m55 = a55*b55;
- m56 = a56*b56;
- m57 = a57*b57;
- m58 = a58*b58;
- m59 = a59*b59;
- m60 = a60*b60;
- m61 = a61*b61;
- m62 = a62*b62;
- m63 = a63*b63;
- m64 = a64*b64;
- m65 = a65*b65;
- m66 = a66*b66;
- m67 = a67*b67;
- m68 = a68*b68;
- m69 = a69*b69;
- m70 = a70*b70;
- m71 = a71*b71;
- m72 = a72*b72;
- m73 = a73*b73;
- m74 = a74*b74;
- m75 = a75*b75;
- m76 = a76*b76;
- m77 = a77*b77;
- m78 = a78*b78;
- m79 = a79*b79;
- m80 = a80*b80;
- m81 = a81*b81;
- m82 = a82*b82;
- m83 = a83*b83;
- m84 = a84*b84;
- m85 = a85*b85;
- m86 = a86*b86;
- m87 = a87*b87;
- m88 = a88*b88;
- m89 = a89*b89;
- m90 = a90*b90;
- m91 = a91*b91;
- m92 = a92*b92;
- m93 = a93*b93;
- m94 = a94*b94;
- m95 = a95*b95;
- m96 = a96*b96;
- m97 = a97*b97;
- m98 = a98*b98;
- m99 = a99*b99;
- m100 = a100*b100;
- m101 = a101*b101;
- m102 = a102*b102;
- m103 = a103*b103;
- m104 = a104*b104;
- m105 = a105*b105;
- m106 = a106*b106;
- m107 = a107*b107;
- m108 = a108*b108;
- m109 = a109*b109;
- m110 = a110*b110;
- m111 = a111*b111;
- m112 = a112*b112;
- m113 = a113*b113;
- m114 = a114*b114;
- m115 = a115*b115;
- m116 = a116*b116;
- m117 = a117*b117;
- m118 = a118*b118;
- m119 = a119*b119;
- m120 = a120*b120;
- m121 = a121*b121;
- m122 = a122*b122;
- m123 = a123*b123;
- m124 = a124*b124;
- m125 = a125*b125;
- m126 = a126*b126;
- m127 = a127*b127;
- m128 = a128*b128;
- m129 = a129*b129;
- m130 = a130*b130;
- m131 = a131*b131;
- m132 = a132*b132;
- m133 = a133*b133;
- m134 = a134*b134;
- m135 = a135*b135;
- m136 = a136*b136;
- m137 = a137*b137;
- m138 = a138*b138;
- m139 = a139*b139;
- m140 = a140*b140;
- m141 = a141*b141;
- m142 = a142*b142;
- m143 = a143*b143;
- m144 = a144*b144;
- m145 = a145*b145;
- m146 = a146*b146;
- m147 = a147*b147;
- m148 = a148*b148;
- m149 = a149*b149;
- m150 = a150*b150;
- m151 = a151*b151;
- m152 = a152*b152;
- m153 = a153*b153;
- m154 = a154*b154;
- m155 = a155*b155;
- m156 = a156*b156;
- m157 = a157*b157;
- m158 = a158*b158;
- m159 = a159*b159;
- m160 = a160*b160;
- m161 = a161*b161;
- m162 = a162*b162;
- m163 = a163*b163;
- m164 = a164*b164;
- m165 = a165*b165;
- m166 = a166*b166;
- m167 = a167*b167;
- m168 = a168*b168;
- m169 = a169*b169;
- m170 = a170*b170;
- m171 = a171*b171;
- m172 = a172*b172;
- m173 = a173*b173;
- m174 = a174*b174;
- m175 = a175*b175;
- m176 = a176*b176;
- m177 = a177*b177;
- m178 = a178*b178;
- m179 = a179*b179;
- m180 = a180*b180;
- m181 = a181*b181;
- m182 = a182*b182;
- m183 = a183*b183;
- m184 = a184*b184;
- m185 = a185*b185;
- m186 = a186*b186;
- m187 = a187*b187;
- m188 = a188*b188;
- m189 = a189*b189;
- m190 = a190*b190;
- m191 = a191*b191;
- m192 = a192*b192;
- m193 = a193*b193;
- m194 = a194*b194;
- m195 = a195*b195;
- m196 = a196*b196;
- m197 = a197*b197;
- m198 = a198*b198;
- m199 = a199*b199;
- m200 = a200*b200;
- m201 = a201*b201;
- m202 = a202*b202;
- m203 = a203*b203;
- m204 = a204*b204;
- m205 = a205*b205;
- m206 = a206*b206;
- m207 = a207*b207;
- m208 = a208*b208;
- m209 = a209*b209;
- m210 = a210*b210;
- m211 = a211*b211;
- m212 = a212*b212;
- m213 = a213*b213;
- m214 = a214*b214;
- m215 = a215*b215;
- m216 = a216*b216;
- m217 = a217*b217;
- m218 = a218*b218;
- m219 = a219*b219;
- m220 = a220*b220;
- m221 = a221*b221;
- m222 = a222*b222;
- m223 = a223*b223;
- m224 = a224*b224;
- m225 = a225*b225;
- m226 = a226*b226;
- m227 = a227*b227;
- m228 = a228*b228;
- m229 = a229*b229;
- m230 = a230*b230;
- m231 = a231*b231;
- m232 = a232*b232;
- m233 = a233*b233;
- m234 = a234*b234;
- m235 = a235*b235;
- m236 = a236*b236;
- m237 = a237*b237;
- m238 = a238*b238;
- m239 = a239*b239;
- m240 = a240*b240;
- m241 = a241*b241;
- m242 = a242*b242;
- m243 = a243*b243;
- m244 = a244*b244;
- m245 = a245*b245;
- m246 = a246*b246;
- m247 = a247*b247;
- m248 = a248*b248;
- m249 = a249*b249;
- m250 = a250*b250;
- m251 = a251*b251;
- m252 = a252*b252;
- m253 = a253*b253;
- m254 = a254*b254;
- m255 = a255*b255;
- m256 = a256*b256;
- add1 = m1+m2;
- add2 = m3+m4;
- add3 = m5+m6;
- add4 = m7+m8;
- add5 = m9+m10;
- add6 = m11+m12;
- add7 = m13+m14;
- add8 = m15+m16;
- add9 = m17+m18;
- add10 = m19+m20;
- add11 = m21+m22;
- add12 = m23+m24;
- add13 = m25+m26;
- add14 = m27+m28;
- add15 = m29+m30;
- add16 = m31+m32;
- add17 = m33+m34;
- add18 = m35+m36;
- add19 = m37+m38;
- add20 = m39+m40;
- add21 = m41+m42;
- add22 = m43+m44;
- add23 = m45+m46;
- add24 = m47+m48;
- add25 = m49+m50;
- add26 = m51+m52;
- add27 = m53+m54;
- add28 = m55+m56;
- add29 = m57+m58;
- add30 = m59+m60;
- add31 = m61+m62;
- add32 = m63+m64;
- add33 = m65+m66;
- add34 = m67+m68;
- add35 = m69+m70;
- add36 = m71+m72;
- add37 = m73+m74;
- add38 = m75+m76;
- add39 = m77+m78;
- add40 = m79+m80;
- add41 = m81+m82;
- add42 = m83+m84;
- add43 = m85+m86;
- add44 = m87+m88;
- add45 = m89+m90;
- add46 = m91+m92;
- add47 = m93+m94;
- add48 = m95+m96;
- add49 = m97+m98;
- add50 = m99+m100;
- add51 = m101+m102;
- add52 = m103+m104;
- add53 = m105+m106;
- add54 = m107+m108;
- add55 = m109+m110;
- add56 = m111+m112;
- add57 = m113+m114;
- add58 = m115+m116;
- add59 = m117+m118;
- add60 = m119+m120;
- add61 = m121+m122;
- add62 = m123+m124;
- add63 = m125+m126;
- add64 = m127+m128;
- add65 = m129+m130;
- add66 = m131+m132;
- add67 = m133+m134;
- add68 = m135+m136;
- add69 = m137+m138;
- add70 = m139+m140;
- add71 = m141+m142;
- add72 = m143+m144;
- add73 = m145+m146;
- add74 = m147+m148;
- add75 = m149+m150;
- add76 = m151+m152;
- add77 = m153+m154;
- add78 = m155+m156;
- add79 = m157+m158;
- add80 = m159+m160;
- add81 = m161+m162;
- add82 = m163+m164;
- add83 = m165+m166;
- add84 = m167+m168;
- add85 = m169+m170;
- add86 = m171+m172;
- add87 = m173+m174;
- add88 = m175+m176;
- add89 = m177+m178;
- add90 = m179+m180;
- add91 = m181+m182;
- add92 = m183+m184;
- add93 = m185+m186;
- add94 = m187+m188;
- add95 = m189+m190;
- add96 = m191+m192;
- add97 = m193+m194;
- add98 = m195+m196;
- add99 = m197+m198;
- add100 = m199+m200;
- add101 = m201+m202;
- add102 = m203+m204;
- add103 = m205+m206;
- add104 = m207+m208;
- add105 = m209+m210;
- add106 = m211+m212;
- add107 = m213+m214;
- add108 = m215+m216;
- add109 = m217+m218;
- add110 = m219+m220;
- add111 = m221+m222;
- add112 = m223+m224;
- add113 = m225+m226;
- add114 = m227+m228;
- add115 = m229+m230;
- add116 = m231+m232;
- add117 = m233+m234;
- add118 = m235+m236;
- add119 = m237+m238;
- add120 = m239+m240;
- add121 = m241+m242;
- add122 = m243+m244;
- add123 = m245+m246;
- add124 = m247+m248;
- add125 = m249+m250;
- add126 = m251+m252;
- add127 = m253+m254;
- add128 = m255+m256;
- add129 = add128+add127;
- add130 = add126+add125;
- add131 = add124+add123;
- add132 = add122+add121;
- add133 = add120+add119;
- add134 = add118+add117;
- add135 = add116+add115;
- add136 = add114+add113;
- add137 = add112+add111;
- add138 = add110+add109;
- add139 = add108+add107;
- add140 = add106+add105;
- add141 = add104+add103;
- add142 = add102+add101;
- add143 = add100+add99;
- add144 = add98+add97;
- add145 = add96+add95;
- add146 = add94+add93;
- add147 = add92+add91;
- add148 = add90+add89;
- add149 = add88+add87;
- add150 = add86+add85;
- add151 = add84+add83;
- add152 = add82+add81;
- add153 = add80+add79;
- add154 = add78+add77;
- add155 = add76+add75;
- add156 = add74+add73;
- add157 = add72+add71;
- add158 = add70+add69;
- add159 = add68+add67;
- add160 = add66+add65;
- add161 = add64+add63;
- add162 = add62+add61;
- add163 = add60+add59;
- add164 = add58+add57;
- add165 = add56+add55;
- add166 = add54+add53;
- add167 = add52+add51;
- add168 = add50+add49;
- add169 = add48+add47;
- add170 = add46+add45;
- add171 = add44+add43;
- add172 = add42+add41;
- add173 = add40+add39;
- add174 = add38+add37;
- add175 = add36+add35;
- add176 = add34+add33;
- add177 = add32+add31;
- add178 = add30+add29;
- add179 = add28+add27;
- add180 = add26+add25;
- add181 = add24+add23;
- add182 = add22+add21;
- add183 = add20+add19;
- add184 = add18+add17;
- add185 = add16+add15;
- add186 = add14+add13;
- add187 = add12+add11;
- add188 = add10+add9;
- add189 = add8+add7;
- add190 = add6+add5;
- add191 = add4+add3;
- add192 = add2+add1;
- add193 = add192+add191;
- add194 = add190+add189;
- add195 = add188+add187;
- add196 = add186+add185;
- add197 = add184+add183;
- add198 = add182+add181;
- add199 = add180+add179;
- add200 = add178+add177;
- add201 = add176+add175;
- add202 = add174+add173;
- add203 = add172+add171;
- add204 = add170+add169;
- add205 = add168+add167;
- add206 = add166+add165;
- add207 = add164+add163;
- add208 = add162+add161;
- add209 = add160+add159;
- add210 = add158+add157;
- add211 = add156+add155;
- add212 = add154+add153;
- add213 = add152+add151;
- add214 = add150+add149;
- add215 = add148+add147;
- add216 = add146+add145;
- add217 = add144+add143;
- add218 = add142+add141;
- add219 = add140+add139;
- add220 = add138+add137;
- add221 = add136+add135;
- add222 = add134+add133;
- add223 = add132+add131;
- add224 = add130+add129;
- add225 = add224+add223;
- add226 = add222+add221;
- add227 = add220+add219;
- add228 = add218+add217;
- add229 = add216+add215;
- add230 = add214+add213;
- add231 = add212+add211;
- add232 = add210+add209;
- add233 = add208+add207;
- add234 = add206+add205;
- add235 = add204+add203;
- add236 = add202+add201;
- add237 = add200+add199;
- add238 = add198+add197;
- add239 = add196+add195;
- add240 = add194+add193;
- add241 = add240+add239;
- add242 = add238+add237;
- add243 = add236+add235;
- add244 = add234+add233;
- add245 = add232+add231;
- add246 = add230+add229;
- add247 = add228+add227;
- add248 = add226+add225;
- add249 = add248+add247;
- add250 = add246+add245;
- add251 = add244+add243;
- add252 = add242+add241;
- add253 = add252+add251;
- add254 = add250+add249;
- return add253+add254;
-}
-float compute_engine512(
-float a1, float b1,
-float a2, float b2,
-float a3, float b3,
-float a4, float b4,
-float a5, float b5,
-float a6, float b6,
-float a7, float b7,
-float a8, float b8,
-float a9, float b9,
-float a10, float b10,
-float a11, float b11,
-float a12, float b12,
-float a13, float b13,
-float a14, float b14,
-float a15, float b15,
-float a16, float b16,
-float a17, float b17,
-float a18, float b18,
-float a19, float b19,
-float a20, float b20,
-float a21, float b21,
-float a22, float b22,
-float a23, float b23,
-float a24, float b24,
-float a25, float b25,
-float a26, float b26,
-float a27, float b27,
-float a28, float b28,
-float a29, float b29,
-float a30, float b30,
-float a31, float b31,
-float a32, float b32,
-float a33, float b33,
-float a34, float b34,
-float a35, float b35,
-float a36, float b36,
-float a37, float b37,
-float a38, float b38,
-float a39, float b39,
-float a40, float b40,
-float a41, float b41,
-float a42, float b42,
-float a43, float b43,
-float a44, float b44,
-float a45, float b45,
-float a46, float b46,
-float a47, float b47,
-float a48, float b48,
-float a49, float b49,
-float a50, float b50,
-float a51, float b51,
-float a52, float b52,
-float a53, float b53,
-float a54, float b54,
-float a55, float b55,
-float a56, float b56,
-float a57, float b57,
-float a58, float b58,
-float a59, float b59,
-float a60, float b60,
-float a61, float b61,
-float a62, float b62,
-float a63, float b63,
-float a64, float b64,
-float a65, float b65,
-float a66, float b66,
-float a67, float b67,
-float a68, float b68,
-float a69, float b69,
-float a70, float b70,
-float a71, float b71,
-float a72, float b72,
-float a73, float b73,
-float a74, float b74,
-float a75, float b75,
-float a76, float b76,
-float a77, float b77,
-float a78, float b78,
-float a79, float b79,
-float a80, float b80,
-float a81, float b81,
-float a82, float b82,
-float a83, float b83,
-float a84, float b84,
-float a85, float b85,
-float a86, float b86,
-float a87, float b87,
-float a88, float b88,
-float a89, float b89,
-float a90, float b90,
-float a91, float b91,
-float a92, float b92,
-float a93, float b93,
-float a94, float b94,
-float a95, float b95,
-float a96, float b96,
-float a97, float b97,
-float a98, float b98,
-float a99, float b99,
-float a100, float b100,
-float a101, float b101,
-float a102, float b102,
-float a103, float b103,
-float a104, float b104,
-float a105, float b105,
-float a106, float b106,
-float a107, float b107,
-float a108, float b108,
-float a109, float b109,
-float a110, float b110,
-float a111, float b111,
-float a112, float b112,
-float a113, float b113,
-float a114, float b114,
-float a115, float b115,
-float a116, float b116,
-float a117, float b117,
-float a118, float b118,
-float a119, float b119,
-float a120, float b120,
-float a121, float b121,
-float a122, float b122,
-float a123, float b123,
-float a124, float b124,
-float a125, float b125,
-float a126, float b126,
-float a127, float b127,
-float a128, float b128,
-float a129, float b129,
-float a130, float b130,
-float a131, float b131,
-float a132, float b132,
-float a133, float b133,
-float a134, float b134,
-float a135, float b135,
-float a136, float b136,
-float a137, float b137,
-float a138, float b138,
-float a139, float b139,
-float a140, float b140,
-float a141, float b141,
-float a142, float b142,
-float a143, float b143,
-float a144, float b144,
-float a145, float b145,
-float a146, float b146,
-float a147, float b147,
-float a148, float b148,
-float a149, float b149,
-float a150, float b150,
-float a151, float b151,
-float a152, float b152,
-float a153, float b153,
-float a154, float b154,
-float a155, float b155,
-float a156, float b156,
-float a157, float b157,
-float a158, float b158,
-float a159, float b159,
-float a160, float b160,
-float a161, float b161,
-float a162, float b162,
-float a163, float b163,
-float a164, float b164,
-float a165, float b165,
-float a166, float b166,
-float a167, float b167,
-float a168, float b168,
-float a169, float b169,
-float a170, float b170,
-float a171, float b171,
-float a172, float b172,
-float a173, float b173,
-float a174, float b174,
-float a175, float b175,
-float a176, float b176,
-float a177, float b177,
-float a178, float b178,
-float a179, float b179,
-float a180, float b180,
-float a181, float b181,
-float a182, float b182,
-float a183, float b183,
-float a184, float b184,
-float a185, float b185,
-float a186, float b186,
-float a187, float b187,
-float a188, float b188,
-float a189, float b189,
-float a190, float b190,
-float a191, float b191,
-float a192, float b192,
-float a193, float b193,
-float a194, float b194,
-float a195, float b195,
-float a196, float b196,
-float a197, float b197,
-float a198, float b198,
-float a199, float b199,
-float a200, float b200,
-float a201, float b201,
-float a202, float b202,
-float a203, float b203,
-float a204, float b204,
-float a205, float b205,
-float a206, float b206,
-float a207, float b207,
-float a208, float b208,
-float a209, float b209,
-float a210, float b210,
-float a211, float b211,
-float a212, float b212,
-float a213, float b213,
-float a214, float b214,
-float a215, float b215,
-float a216, float b216,
-float a217, float b217,
-float a218, float b218,
-float a219, float b219,
-float a220, float b220,
-float a221, float b221,
-float a222, float b222,
-float a223, float b223,
-float a224, float b224,
-float a225, float b225,
-float a226, float b226,
-float a227, float b227,
-float a228, float b228,
-float a229, float b229,
-float a230, float b230,
-float a231, float b231,
-float a232, float b232,
-float a233, float b233,
-float a234, float b234,
-float a235, float b235,
-float a236, float b236,
-float a237, float b237,
-float a238, float b238,
-float a239, float b239,
-float a240, float b240,
-float a241, float b241,
-float a242, float b242,
-float a243, float b243,
-float a244, float b244,
-float a245, float b245,
-float a246, float b246,
-float a247, float b247,
-float a248, float b248,
-float a249, float b249,
-float a250, float b250,
-float a251, float b251,
-float a252, float b252,
-float a253, float b253,
-float a254, float b254,
-float a255, float b255,
-float a256, float b256,
-float a257, float b257,
-float a258, float b258,
-float a259, float b259,
-float a260, float b260,
-float a261, float b261,
-float a262, float b262,
-float a263, float b263,
-float a264, float b264,
-float a265, float b265,
-float a266, float b266,
-float a267, float b267,
-float a268, float b268,
-float a269, float b269,
-float a270, float b270,
-float a271, float b271,
-float a272, float b272,
-float a273, float b273,
-float a274, float b274,
-float a275, float b275,
-float a276, float b276,
-float a277, float b277,
-float a278, float b278,
-float a279, float b279,
-float a280, float b280,
-float a281, float b281,
-float a282, float b282,
-float a283, float b283,
-float a284, float b284,
-float a285, float b285,
-float a286, float b286,
-float a287, float b287,
-float a288, float b288,
-float a289, float b289,
-float a290, float b290,
-float a291, float b291,
-float a292, float b292,
-float a293, float b293,
-float a294, float b294,
-float a295, float b295,
-float a296, float b296,
-float a297, float b297,
-float a298, float b298,
-float a299, float b299,
-float a300, float b300,
-float a301, float b301,
-float a302, float b302,
-float a303, float b303,
-float a304, float b304,
-float a305, float b305,
-float a306, float b306,
-float a307, float b307,
-float a308, float b308,
-float a309, float b309,
-float a310, float b310,
-float a311, float b311,
-float a312, float b312,
-float a313, float b313,
-float a314, float b314,
-float a315, float b315,
-float a316, float b316,
-float a317, float b317,
-float a318, float b318,
-float a319, float b319,
-float a320, float b320,
-float a321, float b321,
-float a322, float b322,
-float a323, float b323,
-float a324, float b324,
-float a325, float b325,
-float a326, float b326,
-float a327, float b327,
-float a328, float b328,
-float a329, float b329,
-float a330, float b330,
-float a331, float b331,
-float a332, float b332,
-float a333, float b333,
-float a334, float b334,
-float a335, float b335,
-float a336, float b336,
-float a337, float b337,
-float a338, float b338,
-float a339, float b339,
-float a340, float b340,
-float a341, float b341,
-float a342, float b342,
-float a343, float b343,
-float a344, float b344,
-float a345, float b345,
-float a346, float b346,
-float a347, float b347,
-float a348, float b348,
-float a349, float b349,
-float a350, float b350,
-float a351, float b351,
-float a352, float b352,
-float a353, float b353,
-float a354, float b354,
-float a355, float b355,
-float a356, float b356,
-float a357, float b357,
-float a358, float b358,
-float a359, float b359,
-float a360, float b360,
-float a361, float b361,
-float a362, float b362,
-float a363, float b363,
-float a364, float b364,
-float a365, float b365,
-float a366, float b366,
-float a367, float b367,
-float a368, float b368,
-float a369, float b369,
-float a370, float b370,
-float a371, float b371,
-float a372, float b372,
-float a373, float b373,
-float a374, float b374,
-float a375, float b375,
-float a376, float b376,
-float a377, float b377,
-float a378, float b378,
-float a379, float b379,
-float a380, float b380,
-float a381, float b381,
-float a382, float b382,
-float a383, float b383,
-float a384, float b384,
-float a385, float b385,
-float a386, float b386,
-float a387, float b387,
-float a388, float b388,
-float a389, float b389,
-float a390, float b390,
-float a391, float b391,
-float a392, float b392,
-float a393, float b393,
-float a394, float b394,
-float a395, float b395,
-float a396, float b396,
-float a397, float b397,
-float a398, float b398,
-float a399, float b399,
-float a400, float b400,
-float a401, float b401,
-float a402, float b402,
-float a403, float b403,
-float a404, float b404,
-float a405, float b405,
-float a406, float b406,
-float a407, float b407,
-float a408, float b408,
-float a409, float b409,
-float a410, float b410,
-float a411, float b411,
-float a412, float b412,
-float a413, float b413,
-float a414, float b414,
-float a415, float b415,
-float a416, float b416,
-float a417, float b417,
-float a418, float b418,
-float a419, float b419,
-float a420, float b420,
-float a421, float b421,
-float a422, float b422,
-float a423, float b423,
-float a424, float b424,
-float a425, float b425,
-float a426, float b426,
-float a427, float b427,
-float a428, float b428,
-float a429, float b429,
-float a430, float b430,
-float a431, float b431,
-float a432, float b432,
-float a433, float b433,
-float a434, float b434,
-float a435, float b435,
-float a436, float b436,
-float a437, float b437,
-float a438, float b438,
-float a439, float b439,
-float a440, float b440,
-float a441, float b441,
-float a442, float b442,
-float a443, float b443,
-float a444, float b444,
-float a445, float b445,
-float a446, float b446,
-float a447, float b447,
-float a448, float b448,
-float a449, float b449,
-float a450, float b450,
-float a451, float b451,
-float a452, float b452,
-float a453, float b453,
-float a454, float b454,
-float a455, float b455,
-float a456, float b456,
-float a457, float b457,
-float a458, float b458,
-float a459, float b459,
-float a460, float b460,
-float a461, float b461,
-float a462, float b462,
-float a463, float b463,
-float a464, float b464,
-float a465, float b465,
-float a466, float b466,
-float a467, float b467,
-float a468, float b468,
-float a469, float b469,
-float a470, float b470,
-float a471, float b471,
-float a472, float b472,
-float a473, float b473,
-float a474, float b474,
-float a475, float b475,
-float a476, float b476,
-float a477, float b477,
-float a478, float b478,
-float a479, float b479,
-float a480, float b480,
-float a481, float b481,
-float a482, float b482,
-float a483, float b483,
-float a484, float b484,
-float a485, float b485,
-float a486, float b486,
-float a487, float b487,
-float a488, float b488,
-float a489, float b489,
-float a490, float b490,
-float a491, float b491,
-float a492, float b492,
-float a493, float b493,
-float a494, float b494,
-float a495, float b495,
-float a496, float b496,
-float a497, float b497,
-float a498, float b498,
-float a499, float b499,
-float a500, float b500,
-float a501, float b501,
-float a502, float b502,
-float a503, float b503,
-float a504, float b504,
-float a505, float b505,
-float a506, float b506,
-float a507, float b507,
-float a508, float b508,
-float a509, float b509,
-float a510, float b510,
-float a511, float b511,
-float a512, float b512){
-#pragma HLS PIPELINE
- float m1, m2, m3, m4,
- m5, m6, m7, m8, m9,
- m10, m11, m12, m13, m14,
- m15, m16, m17, m18, m19,
- m20, m21, m22, m23, m24,
- m25, m26, m27, m28, m29,
- m30, m31, m32, m33, m34,
- m35, m36, m37, m38, m39,
- m40, m41, m42, m43, m44,
- m45, m46, m47, m48, m49,
- m50, m51, m52, m53, m54,
- m55, m56, m57, m58, m59,
- m60, m61, m62, m63, m64,
- m65, m66, m67, m68, m69,
- m70, m71, m72, m73, m74,
- m75, m76, m77, m78, m79,
- m80, m81, m82, m83, m84,
- m85, m86, m87, m88, m89,
- m90, m91, m92, m93, m94,
- m95, m96, m97, m98, m99,
- m100, m101, m102, m103, m104,
- m105, m106, m107, m108, m109,
- m110, m111, m112, m113, m114,
- m115, m116, m117, m118, m119,
- m120, m121, m122, m123, m124,
- m125, m126, m127, m128, m129,
- m130, m131, m132, m133, m134,
- m135, m136, m137, m138, m139,
- m140, m141, m142, m143, m144,
- m145, m146, m147, m148, m149,
- m150, m151, m152, m153, m154,
- m155, m156, m157, m158, m159,
- m160, m161, m162, m163, m164,
- m165, m166, m167, m168, m169,
- m170, m171, m172, m173, m174,
- m175, m176, m177, m178, m179,
- m180, m181, m182, m183, m184,
- m185, m186, m187, m188, m189,
- m190, m191, m192, m193, m194,
- m195, m196, m197, m198, m199,
- m200, m201, m202, m203, m204,
- m205, m206, m207, m208, m209,
- m210, m211, m212, m213, m214,
- m215, m216, m217, m218, m219,
- m220, m221, m222, m223, m224,
- m225, m226, m227, m228, m229,
- m230, m231, m232, m233, m234,
- m235, m236, m237, m238, m239,
- m240, m241, m242, m243, m244,
- m245, m246, m247, m248, m249,
- m250, m251, m252, m253, m254,
- m255, m256, m257, m258, m259,
- m260, m261, m262, m263, m264,
- m265, m266, m267, m268, m269,
- m270, m271, m272, m273, m274,
- m275, m276, m277, m278, m279,
- m280, m281, m282, m283, m284,
- m285, m286, m287, m288, m289,
- m290, m291, m292, m293, m294,
- m295, m296, m297, m298, m299,
- m300, m301, m302, m303, m304,
- m305, m306, m307, m308, m309,
- m310, m311, m312, m313, m314,
- m315, m316, m317, m318, m319,
- m320, m321, m322, m323, m324,
- m325, m326, m327, m328, m329,
- m330, m331, m332, m333, m334,
- m335, m336, m337, m338, m339,
- m340, m341, m342, m343, m344,
- m345, m346, m347, m348, m349,
- m350, m351, m352, m353, m354,
- m355, m356, m357, m358, m359,
- m360, m361, m362, m363, m364,
- m365, m366, m367, m368, m369,
- m370, m371, m372, m373, m374,
- m375, m376, m377, m378, m379,
- m380, m381, m382, m383, m384,
- m385, m386, m387, m388, m389,
- m390, m391, m392, m393, m394,
- m395, m396, m397, m398, m399,
- m400, m401, m402, m403, m404,
- m405, m406, m407, m408, m409,
- m410, m411, m412, m413, m414,
- m415, m416, m417, m418, m419,
- m420, m421, m422, m423, m424,
- m425, m426, m427, m428, m429,
- m430, m431, m432, m433, m434,
- m435, m436, m437, m438, m439,
- m440, m441, m442, m443, m444,
- m445, m446, m447, m448, m449,
- m450, m451, m452, m453, m454,
- m455, m456, m457, m458, m459,
- m460, m461, m462, m463, m464,
- m465, m466, m467, m468, m469,
- m470, m471, m472, m473, m474,
- m475, m476, m477, m478, m479,
- m480, m481, m482, m483, m484,
- m485, m486, m487, m488, m489,
- m490, m491, m492, m493, m494,
- m495, m496, m497, m498, m499,
- m500, m501, m502, m503, m504,
- m505, m506, m507, m508, m509,
- m510, m511, m512;
-
- float add1, add2, add3, add4,
- add5, add6, add7, add8, add9,
- add10, add11, add12, add13, add14,
- add15, add16, add17, add18, add19,
- add20, add21, add22, add23, add24,
- add25, add26, add27, add28, add29,
- add30, add31, add32, add33, add34,
- add35, add36, add37, add38, add39,
- add40, add41, add42, add43, add44,
- add45, add46, add47, add48, add49,
- add50, add51, add52, add53, add54,
- add55, add56, add57, add58, add59,
- add60, add61, add62, add63, add64,
- add65, add66, add67, add68, add69,
- add70, add71, add72, add73, add74,
- add75, add76, add77, add78, add79,
- add80, add81, add82, add83, add84,
- add85, add86, add87, add88, add89,
- add90, add91, add92, add93, add94,
- add95, add96, add97, add98, add99,
- add100, add101, add102, add103, add104,
- add105, add106, add107, add108, add109,
- add110, add111, add112, add113, add114,
- add115, add116, add117, add118, add119,
- add120, add121, add122, add123, add124,
- add125, add126, add127, add128, add129,
- add130, add131, add132, add133, add134,
- add135, add136, add137, add138, add139,
- add140, add141, add142, add143, add144,
- add145, add146, add147, add148, add149,
- add150, add151, add152, add153, add154,
- add155, add156, add157, add158, add159,
- add160, add161, add162, add163, add164,
- add165, add166, add167, add168, add169,
- add170, add171, add172, add173, add174,
- add175, add176, add177, add178, add179,
- add180, add181, add182, add183, add184,
- add185, add186, add187, add188, add189,
- add190, add191, add192, add193, add194,
- add195, add196, add197, add198, add199,
- add200, add201, add202, add203, add204,
- add205, add206, add207, add208, add209,
- add210, add211, add212, add213, add214,
- add215, add216, add217, add218, add219,
- add220, add221, add222, add223, add224,
- add225, add226, add227, add228, add229,
- add230, add231, add232, add233, add234,
- add235, add236, add237, add238, add239,
- add240, add241, add242, add243, add244,
- add245, add246, add247, add248, add249,
- add250, add251, add252, add253, add254,
- add255, add256, add257, add258, add259,
- add260, add261, add262, add263, add264,
- add265, add266, add267, add268, add269,
- add270, add271, add272, add273, add274,
- add275, add276, add277, add278, add279,
- add280, add281, add282, add283, add284,
- add285, add286, add287, add288, add289,
- add290, add291, add292, add293, add294,
- add295, add296, add297, add298, add299,
- add300, add301, add302, add303, add304,
- add305, add306, add307, add308, add309,
- add310, add311, add312, add313, add314,
- add315, add316, add317, add318, add319,
- add320, add321, add322, add323, add324,
- add325, add326, add327, add328, add329,
- add330, add331, add332, add333, add334,
- add335, add336, add337, add338, add339,
- add340, add341, add342, add343, add344,
- add345, add346, add347, add348, add349,
- add350, add351, add352, add353, add354,
- add355, add356, add357, add358, add359,
- add360, add361, add362, add363, add364,
- add365, add366, add367, add368, add369,
- add370, add371, add372, add373, add374,
- add375, add376, add377, add378, add379,
- add380, add381, add382, add383, add384,
- add385, add386, add387, add388, add389,
- add390, add391, add392, add393, add394,
- add395, add396, add397, add398, add399,
- add400, add401, add402, add403, add404,
- add405, add406, add407, add408, add409,
- add410, add411, add412, add413, add414,
- add415, add416, add417, add418, add419,
- add420, add421, add422, add423, add424,
- add425, add426, add427, add428, add429,
- add430, add431, add432, add433, add434,
- add435, add436, add437, add438, add439,
- add440, add441, add442, add443, add444,
- add445, add446, add447, add448, add449,
- add450, add451, add452, add453, add454,
- add455, add456, add457, add458, add459,
- add460, add461, add462, add463, add464,
- add465, add466, add467, add468, add469,
- add470, add471, add472, add473, add474,
- add475, add476, add477, add478, add479,
- add480, add481, add482, add483, add484,
- add485, add486, add487, add488, add489,
- add490, add491, add492, add493, add494,
- add495, add496, add497, add498, add499,
- add500, add501, add502, add503, add504,
- add505, add506, add507, add508, add509,
- add510;
- m1 = a1*b1;
- m2 = a2*b2;
- m3 = a3*b3;
- m4 = a4*b4;
- m5 = a5*b5;
- m6 = a6*b6;
- m7 = a7*b7;
- m8 = a8*b8;
- m9 = a9*b9;
- m10 = a10*b10;
- m11 = a11*b11;
- m12 = a12*b12;
- m13 = a13*b13;
- m14 = a14*b14;
- m15 = a15*b15;
- m16 = a16*b16;
- m17 = a17*b17;
- m18 = a18*b18;
- m19 = a19*b19;
- m20 = a20*b20;
- m21 = a21*b21;
- m22 = a22*b22;
- m23 = a23*b23;
- m24 = a24*b24;
- m25 = a25*b25;
- m26 = a26*b26;
- m27 = a27*b27;
- m28 = a28*b28;
- m29 = a29*b29;
- m30 = a30*b30;
- m31 = a31*b31;
- m32 = a32*b32;
- m33 = a33*b33;
- m34 = a34*b34;
- m35 = a35*b35;
- m36 = a36*b36;
- m37 = a37*b37;
- m38 = a38*b38;
- m39 = a39*b39;
- m40 = a40*b40;
- m41 = a41*b41;
- m42 = a42*b42;
- m43 = a43*b43;
- m44 = a44*b44;
- m45 = a45*b45;
- m46 = a46*b46;
- m47 = a47*b47;
- m48 = a48*b48;
- m49 = a49*b49;
- m50 = a50*b50;
- m51 = a51*b51;
- m52 = a52*b52;
- m53 = a53*b53;
- m54 = a54*b54;
- m55 = a55*b55;
- m56 = a56*b56;
- m57 = a57*b57;
- m58 = a58*b58;
- m59 = a59*b59;
- m60 = a60*b60;
- m61 = a61*b61;
- m62 = a62*b62;
- m63 = a63*b63;
- m64 = a64*b64;
- m65 = a65*b65;
- m66 = a66*b66;
- m67 = a67*b67;
- m68 = a68*b68;
- m69 = a69*b69;
- m70 = a70*b70;
- m71 = a71*b71;
- m72 = a72*b72;
- m73 = a73*b73;
- m74 = a74*b74;
- m75 = a75*b75;
- m76 = a76*b76;
- m77 = a77*b77;
- m78 = a78*b78;
- m79 = a79*b79;
- m80 = a80*b80;
- m81 = a81*b81;
- m82 = a82*b82;
- m83 = a83*b83;
- m84 = a84*b84;
- m85 = a85*b85;
- m86 = a86*b86;
- m87 = a87*b87;
- m88 = a88*b88;
- m89 = a89*b89;
- m90 = a90*b90;
- m91 = a91*b91;
- m92 = a92*b92;
- m93 = a93*b93;
- m94 = a94*b94;
- m95 = a95*b95;
- m96 = a96*b96;
- m97 = a97*b97;
- m98 = a98*b98;
- m99 = a99*b99;
- m100 = a100*b100;
- m101 = a101*b101;
- m102 = a102*b102;
- m103 = a103*b103;
- m104 = a104*b104;
- m105 = a105*b105;
- m106 = a106*b106;
- m107 = a107*b107;
- m108 = a108*b108;
- m109 = a109*b109;
- m110 = a110*b110;
- m111 = a111*b111;
- m112 = a112*b112;
- m113 = a113*b113;
- m114 = a114*b114;
- m115 = a115*b115;
- m116 = a116*b116;
- m117 = a117*b117;
- m118 = a118*b118;
- m119 = a119*b119;
- m120 = a120*b120;
- m121 = a121*b121;
- m122 = a122*b122;
- m123 = a123*b123;
- m124 = a124*b124;
- m125 = a125*b125;
- m126 = a126*b126;
- m127 = a127*b127;
- m128 = a128*b128;
- m129 = a129*b129;
- m130 = a130*b130;
- m131 = a131*b131;
- m132 = a132*b132;
- m133 = a133*b133;
- m134 = a134*b134;
- m135 = a135*b135;
- m136 = a136*b136;
- m137 = a137*b137;
- m138 = a138*b138;
- m139 = a139*b139;
- m140 = a140*b140;
- m141 = a141*b141;
- m142 = a142*b142;
- m143 = a143*b143;
- m144 = a144*b144;
- m145 = a145*b145;
- m146 = a146*b146;
- m147 = a147*b147;
- m148 = a148*b148;
- m149 = a149*b149;
- m150 = a150*b150;
- m151 = a151*b151;
- m152 = a152*b152;
- m153 = a153*b153;
- m154 = a154*b154;
- m155 = a155*b155;
- m156 = a156*b156;
- m157 = a157*b157;
- m158 = a158*b158;
- m159 = a159*b159;
- m160 = a160*b160;
- m161 = a161*b161;
- m162 = a162*b162;
- m163 = a163*b163;
- m164 = a164*b164;
- m165 = a165*b165;
- m166 = a166*b166;
- m167 = a167*b167;
- m168 = a168*b168;
- m169 = a169*b169;
- m170 = a170*b170;
- m171 = a171*b171;
- m172 = a172*b172;
- m173 = a173*b173;
- m174 = a174*b174;
- m175 = a175*b175;
- m176 = a176*b176;
- m177 = a177*b177;
- m178 = a178*b178;
- m179 = a179*b179;
- m180 = a180*b180;
- m181 = a181*b181;
- m182 = a182*b182;
- m183 = a183*b183;
- m184 = a184*b184;
- m185 = a185*b185;
- m186 = a186*b186;
- m187 = a187*b187;
- m188 = a188*b188;
- m189 = a189*b189;
- m190 = a190*b190;
- m191 = a191*b191;
- m192 = a192*b192;
- m193 = a193*b193;
- m194 = a194*b194;
- m195 = a195*b195;
- m196 = a196*b196;
- m197 = a197*b197;
- m198 = a198*b198;
- m199 = a199*b199;
- m200 = a200*b200;
- m201 = a201*b201;
- m202 = a202*b202;
- m203 = a203*b203;
- m204 = a204*b204;
- m205 = a205*b205;
- m206 = a206*b206;
- m207 = a207*b207;
- m208 = a208*b208;
- m209 = a209*b209;
- m210 = a210*b210;
- m211 = a211*b211;
- m212 = a212*b212;
- m213 = a213*b213;
- m214 = a214*b214;
- m215 = a215*b215;
- m216 = a216*b216;
- m217 = a217*b217;
- m218 = a218*b218;
- m219 = a219*b219;
- m220 = a220*b220;
- m221 = a221*b221;
- m222 = a222*b222;
- m223 = a223*b223;
- m224 = a224*b224;
- m225 = a225*b225;
- m226 = a226*b226;
- m227 = a227*b227;
- m228 = a228*b228;
- m229 = a229*b229;
- m230 = a230*b230;
- m231 = a231*b231;
- m232 = a232*b232;
- m233 = a233*b233;
- m234 = a234*b234;
- m235 = a235*b235;
- m236 = a236*b236;
- m237 = a237*b237;
- m238 = a238*b238;
- m239 = a239*b239;
- m240 = a240*b240;
- m241 = a241*b241;
- m242 = a242*b242;
- m243 = a243*b243;
- m244 = a244*b244;
- m245 = a245*b245;
- m246 = a246*b246;
- m247 = a247*b247;
- m248 = a248*b248;
- m249 = a249*b249;
- m250 = a250*b250;
- m251 = a251*b251;
- m252 = a252*b252;
- m253 = a253*b253;
- m254 = a254*b254;
- m255 = a255*b255;
- m256 = a256*b256;
- m257 = a257*b257;
- m258 = a258*b258;
- m259 = a259*b259;
- m260 = a260*b260;
- m261 = a261*b261;
- m262 = a262*b262;
- m263 = a263*b263;
- m264 = a264*b264;
- m265 = a265*b265;
- m266 = a266*b266;
- m267 = a267*b267;
- m268 = a268*b268;
- m269 = a269*b269;
- m270 = a270*b270;
- m271 = a271*b271;
- m272 = a272*b272;
- m273 = a273*b273;
- m274 = a274*b274;
- m275 = a275*b275;
- m276 = a276*b276;
- m277 = a277*b277;
- m278 = a278*b278;
- m279 = a279*b279;
- m280 = a280*b280;
- m281 = a281*b281;
- m282 = a282*b282;
- m283 = a283*b283;
- m284 = a284*b284;
- m285 = a285*b285;
- m286 = a286*b286;
- m287 = a287*b287;
- m288 = a288*b288;
- m289 = a289*b289;
- m290 = a290*b290;
- m291 = a291*b291;
- m292 = a292*b292;
- m293 = a293*b293;
- m294 = a294*b294;
- m295 = a295*b295;
- m296 = a296*b296;
- m297 = a297*b297;
- m298 = a298*b298;
- m299 = a299*b299;
- m300 = a300*b300;
- m301 = a301*b301;
- m302 = a302*b302;
- m303 = a303*b303;
- m304 = a304*b304;
- m305 = a305*b305;
- m306 = a306*b306;
- m307 = a307*b307;
- m308 = a308*b308;
- m309 = a309*b309;
- m310 = a310*b310;
- m311 = a311*b311;
- m312 = a312*b312;
- m313 = a313*b313;
- m314 = a314*b314;
- m315 = a315*b315;
- m316 = a316*b316;
- m317 = a317*b317;
- m318 = a318*b318;
- m319 = a319*b319;
- m320 = a320*b320;
- m321 = a321*b321;
- m322 = a322*b322;
- m323 = a323*b323;
- m324 = a324*b324;
- m325 = a325*b325;
- m326 = a326*b326;
- m327 = a327*b327;
- m328 = a328*b328;
- m329 = a329*b329;
- m330 = a330*b330;
- m331 = a331*b331;
- m332 = a332*b332;
- m333 = a333*b333;
- m334 = a334*b334;
- m335 = a335*b335;
- m336 = a336*b336;
- m337 = a337*b337;
- m338 = a338*b338;
- m339 = a339*b339;
- m340 = a340*b340;
- m341 = a341*b341;
- m342 = a342*b342;
- m343 = a343*b343;
- m344 = a344*b344;
- m345 = a345*b345;
- m346 = a346*b346;
- m347 = a347*b347;
- m348 = a348*b348;
- m349 = a349*b349;
- m350 = a350*b350;
- m351 = a351*b351;
- m352 = a352*b352;
- m353 = a353*b353;
- m354 = a354*b354;
- m355 = a355*b355;
- m356 = a356*b356;
- m357 = a357*b357;
- m358 = a358*b358;
- m359 = a359*b359;
- m360 = a360*b360;
- m361 = a361*b361;
- m362 = a362*b362;
- m363 = a363*b363;
- m364 = a364*b364;
- m365 = a365*b365;
- m366 = a366*b366;
- m367 = a367*b367;
- m368 = a368*b368;
- m369 = a369*b369;
- m370 = a370*b370;
- m371 = a371*b371;
- m372 = a372*b372;
- m373 = a373*b373;
- m374 = a374*b374;
- m375 = a375*b375;
- m376 = a376*b376;
- m377 = a377*b377;
- m378 = a378*b378;
- m379 = a379*b379;
- m380 = a380*b380;
- m381 = a381*b381;
- m382 = a382*b382;
- m383 = a383*b383;
- m384 = a384*b384;
- m385 = a385*b385;
- m386 = a386*b386;
- m387 = a387*b387;
- m388 = a388*b388;
- m389 = a389*b389;
- m390 = a390*b390;
- m391 = a391*b391;
- m392 = a392*b392;
- m393 = a393*b393;
- m394 = a394*b394;
- m395 = a395*b395;
- m396 = a396*b396;
- m397 = a397*b397;
- m398 = a398*b398;
- m399 = a399*b399;
- m400 = a400*b400;
- m401 = a401*b401;
- m402 = a402*b402;
- m403 = a403*b403;
- m404 = a404*b404;
- m405 = a405*b405;
- m406 = a406*b406;
- m407 = a407*b407;
- m408 = a408*b408;
- m409 = a409*b409;
- m410 = a410*b410;
- m411 = a411*b411;
- m412 = a412*b412;
- m413 = a413*b413;
- m414 = a414*b414;
- m415 = a415*b415;
- m416 = a416*b416;
- m417 = a417*b417;
- m418 = a418*b418;
- m419 = a419*b419;
- m420 = a420*b420;
- m421 = a421*b421;
- m422 = a422*b422;
- m423 = a423*b423;
- m424 = a424*b424;
- m425 = a425*b425;
- m426 = a426*b426;
- m427 = a427*b427;
- m428 = a428*b428;
- m429 = a429*b429;
- m430 = a430*b430;
- m431 = a431*b431;
- m432 = a432*b432;
- m433 = a433*b433;
- m434 = a434*b434;
- m435 = a435*b435;
- m436 = a436*b436;
- m437 = a437*b437;
- m438 = a438*b438;
- m439 = a439*b439;
- m440 = a440*b440;
- m441 = a441*b441;
- m442 = a442*b442;
- m443 = a443*b443;
- m444 = a444*b444;
- m445 = a445*b445;
- m446 = a446*b446;
- m447 = a447*b447;
- m448 = a448*b448;
- m449 = a449*b449;
- m450 = a450*b450;
- m451 = a451*b451;
- m452 = a452*b452;
- m453 = a453*b453;
- m454 = a454*b454;
- m455 = a455*b455;
- m456 = a456*b456;
- m457 = a457*b457;
- m458 = a458*b458;
- m459 = a459*b459;
- m460 = a460*b460;
- m461 = a461*b461;
- m462 = a462*b462;
- m463 = a463*b463;
- m464 = a464*b464;
- m465 = a465*b465;
- m466 = a466*b466;
- m467 = a467*b467;
- m468 = a468*b468;
- m469 = a469*b469;
- m470 = a470*b470;
- m471 = a471*b471;
- m472 = a472*b472;
- m473 = a473*b473;
- m474 = a474*b474;
- m475 = a475*b475;
- m476 = a476*b476;
- m477 = a477*b477;
- m478 = a478*b478;
- m479 = a479*b479;
- m480 = a480*b480;
- m481 = a481*b481;
- m482 = a482*b482;
- m483 = a483*b483;
- m484 = a484*b484;
- m485 = a485*b485;
- m486 = a486*b486;
- m487 = a487*b487;
- m488 = a488*b488;
- m489 = a489*b489;
- m490 = a490*b490;
- m491 = a491*b491;
- m492 = a492*b492;
- m493 = a493*b493;
- m494 = a494*b494;
- m495 = a495*b495;
- m496 = a496*b496;
- m497 = a497*b497;
- m498 = a498*b498;
- m499 = a499*b499;
- m500 = a500*b500;
- m501 = a501*b501;
- m502 = a502*b502;
- m503 = a503*b503;
- m504 = a504*b504;
- m505 = a505*b505;
- m506 = a506*b506;
- m507 = a507*b507;
- m508 = a508*b508;
- m509 = a509*b509;
- m510 = a510*b510;
- m511 = a511*b511;
- m512 = a512*b512;
- add1 = m1+m2;
- add2 = m3+m4;
- add3 = m5+m6;
- add4 = m7+m8;
- add5 = m9+m10;
- add6 = m11+m12;
- add7 = m13+m14;
- add8 = m15+m16;
- add9 = m17+m18;
- add10 = m19+m20;
- add11 = m21+m22;
- add12 = m23+m24;
- add13 = m25+m26;
- add14 = m27+m28;
- add15 = m29+m30;
- add16 = m31+m32;
- add17 = m33+m34;
- add18 = m35+m36;
- add19 = m37+m38;
- add20 = m39+m40;
- add21 = m41+m42;
- add22 = m43+m44;
- add23 = m45+m46;
- add24 = m47+m48;
- add25 = m49+m50;
- add26 = m51+m52;
- add27 = m53+m54;
- add28 = m55+m56;
- add29 = m57+m58;
- add30 = m59+m60;
- add31 = m61+m62;
- add32 = m63+m64;
- add33 = m65+m66;
- add34 = m67+m68;
- add35 = m69+m70;
- add36 = m71+m72;
- add37 = m73+m74;
- add38 = m75+m76;
- add39 = m77+m78;
- add40 = m79+m80;
- add41 = m81+m82;
- add42 = m83+m84;
- add43 = m85+m86;
- add44 = m87+m88;
- add45 = m89+m90;
- add46 = m91+m92;
- add47 = m93+m94;
- add48 = m95+m96;
- add49 = m97+m98;
- add50 = m99+m100;
- add51 = m101+m102;
- add52 = m103+m104;
- add53 = m105+m106;
- add54 = m107+m108;
- add55 = m109+m110;
- add56 = m111+m112;
- add57 = m113+m114;
- add58 = m115+m116;
- add59 = m117+m118;
- add60 = m119+m120;
- add61 = m121+m122;
- add62 = m123+m124;
- add63 = m125+m126;
- add64 = m127+m128;
- add65 = m129+m130;
- add66 = m131+m132;
- add67 = m133+m134;
- add68 = m135+m136;
- add69 = m137+m138;
- add70 = m139+m140;
- add71 = m141+m142;
- add72 = m143+m144;
- add73 = m145+m146;
- add74 = m147+m148;
- add75 = m149+m150;
- add76 = m151+m152;
- add77 = m153+m154;
- add78 = m155+m156;
- add79 = m157+m158;
- add80 = m159+m160;
- add81 = m161+m162;
- add82 = m163+m164;
- add83 = m165+m166;
- add84 = m167+m168;
- add85 = m169+m170;
- add86 = m171+m172;
- add87 = m173+m174;
- add88 = m175+m176;
- add89 = m177+m178;
- add90 = m179+m180;
- add91 = m181+m182;
- add92 = m183+m184;
- add93 = m185+m186;
- add94 = m187+m188;
- add95 = m189+m190;
- add96 = m191+m192;
- add97 = m193+m194;
- add98 = m195+m196;
- add99 = m197+m198;
- add100 = m199+m200;
- add101 = m201+m202;
- add102 = m203+m204;
- add103 = m205+m206;
- add104 = m207+m208;
- add105 = m209+m210;
- add106 = m211+m212;
- add107 = m213+m214;
- add108 = m215+m216;
- add109 = m217+m218;
- add110 = m219+m220;
- add111 = m221+m222;
- add112 = m223+m224;
- add113 = m225+m226;
- add114 = m227+m228;
- add115 = m229+m230;
- add116 = m231+m232;
- add117 = m233+m234;
- add118 = m235+m236;
- add119 = m237+m238;
- add120 = m239+m240;
- add121 = m241+m242;
- add122 = m243+m244;
- add123 = m245+m246;
- add124 = m247+m248;
- add125 = m249+m250;
- add126 = m251+m252;
- add127 = m253+m254;
- add128 = m255+m256;
- add129 = m257+m258;
- add130 = m259+m260;
- add131 = m261+m262;
- add132 = m263+m264;
- add133 = m265+m266;
- add134 = m267+m268;
- add135 = m269+m270;
- add136 = m271+m272;
- add137 = m273+m274;
- add138 = m275+m276;
- add139 = m277+m278;
- add140 = m279+m280;
- add141 = m281+m282;
- add142 = m283+m284;
- add143 = m285+m286;
- add144 = m287+m288;
- add145 = m289+m290;
- add146 = m291+m292;
- add147 = m293+m294;
- add148 = m295+m296;
- add149 = m297+m298;
- add150 = m299+m300;
- add151 = m301+m302;
- add152 = m303+m304;
- add153 = m305+m306;
- add154 = m307+m308;
- add155 = m309+m310;
- add156 = m311+m312;
- add157 = m313+m314;
- add158 = m315+m316;
- add159 = m317+m318;
- add160 = m319+m320;
- add161 = m321+m322;
- add162 = m323+m324;
- add163 = m325+m326;
- add164 = m327+m328;
- add165 = m329+m330;
- add166 = m331+m332;
- add167 = m333+m334;
- add168 = m335+m336;
- add169 = m337+m338;
- add170 = m339+m340;
- add171 = m341+m342;
- add172 = m343+m344;
- add173 = m345+m346;
- add174 = m347+m348;
- add175 = m349+m350;
- add176 = m351+m352;
- add177 = m353+m354;
- add178 = m355+m356;
- add179 = m357+m358;
- add180 = m359+m360;
- add181 = m361+m362;
- add182 = m363+m364;
- add183 = m365+m366;
- add184 = m367+m368;
- add185 = m369+m370;
- add186 = m371+m372;
- add187 = m373+m374;
- add188 = m375+m376;
- add189 = m377+m378;
- add190 = m379+m380;
- add191 = m381+m382;
- add192 = m383+m384;
- add193 = m385+m386;
- add194 = m387+m388;
- add195 = m389+m390;
- add196 = m391+m392;
- add197 = m393+m394;
- add198 = m395+m396;
- add199 = m397+m398;
- add200 = m399+m400;
- add201 = m401+m402;
- add202 = m403+m404;
- add203 = m405+m406;
- add204 = m407+m408;
- add205 = m409+m410;
- add206 = m411+m412;
- add207 = m413+m414;
- add208 = m415+m416;
- add209 = m417+m418;
- add210 = m419+m420;
- add211 = m421+m422;
- add212 = m423+m424;
- add213 = m425+m426;
- add214 = m427+m428;
- add215 = m429+m430;
- add216 = m431+m432;
- add217 = m433+m434;
- add218 = m435+m436;
- add219 = m437+m438;
- add220 = m439+m440;
- add221 = m441+m442;
- add222 = m443+m444;
- add223 = m445+m446;
- add224 = m447+m448;
- add225 = m449+m450;
- add226 = m451+m452;
- add227 = m453+m454;
- add228 = m455+m456;
- add229 = m457+m458;
- add230 = m459+m460;
- add231 = m461+m462;
- add232 = m463+m464;
- add233 = m465+m466;
- add234 = m467+m468;
- add235 = m469+m470;
- add236 = m471+m472;
- add237 = m473+m474;
- add238 = m475+m476;
- add239 = m477+m478;
- add240 = m479+m480;
- add241 = m481+m482;
- add242 = m483+m484;
- add243 = m485+m486;
- add244 = m487+m488;
- add245 = m489+m490;
- add246 = m491+m492;
- add247 = m493+m494;
- add248 = m495+m496;
- add249 = m497+m498;
- add250 = m499+m500;
- add251 = m501+m502;
- add252 = m503+m504;
- add253 = m505+m506;
- add254 = m507+m508;
- add255 = m509+m510;
- add256 = m511+m512;
- add257 = add256+add255;
- add258 = add254+add253;
- add259 = add252+add251;
- add260 = add250+add249;
- add261 = add248+add247;
- add262 = add246+add245;
- add263 = add244+add243;
- add264 = add242+add241;
- add265 = add240+add239;
- add266 = add238+add237;
- add267 = add236+add235;
- add268 = add234+add233;
- add269 = add232+add231;
- add270 = add230+add229;
- add271 = add228+add227;
- add272 = add226+add225;
- add273 = add224+add223;
- add274 = add222+add221;
- add275 = add220+add219;
- add276 = add218+add217;
- add277 = add216+add215;
- add278 = add214+add213;
- add279 = add212+add211;
- add280 = add210+add209;
- add281 = add208+add207;
- add282 = add206+add205;
- add283 = add204+add203;
- add284 = add202+add201;
- add285 = add200+add199;
- add286 = add198+add197;
- add287 = add196+add195;
- add288 = add194+add193;
- add289 = add192+add191;
- add290 = add190+add189;
- add291 = add188+add187;
- add292 = add186+add185;
- add293 = add184+add183;
- add294 = add182+add181;
- add295 = add180+add179;
- add296 = add178+add177;
- add297 = add176+add175;
- add298 = add174+add173;
- add299 = add172+add171;
- add300 = add170+add169;
- add301 = add168+add167;
- add302 = add166+add165;
- add303 = add164+add163;
- add304 = add162+add161;
- add305 = add160+add159;
- add306 = add158+add157;
- add307 = add156+add155;
- add308 = add154+add153;
- add309 = add152+add151;
- add310 = add150+add149;
- add311 = add148+add147;
- add312 = add146+add145;
- add313 = add144+add143;
- add314 = add142+add141;
- add315 = add140+add139;
- add316 = add138+add137;
- add317 = add136+add135;
- add318 = add134+add133;
- add319 = add132+add131;
- add320 = add130+add129;
- add321 = add128+add127;
- add322 = add126+add125;
- add323 = add124+add123;
- add324 = add122+add121;
- add325 = add120+add119;
- add326 = add118+add117;
- add327 = add116+add115;
- add328 = add114+add113;
- add329 = add112+add111;
- add330 = add110+add109;
- add331 = add108+add107;
- add332 = add106+add105;
- add333 = add104+add103;
- add334 = add102+add101;
- add335 = add100+add99;
- add336 = add98+add97;
- add337 = add96+add95;
- add338 = add94+add93;
- add339 = add92+add91;
- add340 = add90+add89;
- add341 = add88+add87;
- add342 = add86+add85;
- add343 = add84+add83;
- add344 = add82+add81;
- add345 = add80+add79;
- add346 = add78+add77;
- add347 = add76+add75;
- add348 = add74+add73;
- add349 = add72+add71;
- add350 = add70+add69;
- add351 = add68+add67;
- add352 = add66+add65;
- add353 = add64+add63;
- add354 = add62+add61;
- add355 = add60+add59;
- add356 = add58+add57;
- add357 = add56+add55;
- add358 = add54+add53;
- add359 = add52+add51;
- add360 = add50+add49;
- add361 = add48+add47;
- add362 = add46+add45;
- add363 = add44+add43;
- add364 = add42+add41;
- add365 = add40+add39;
- add366 = add38+add37;
- add367 = add36+add35;
- add368 = add34+add33;
- add369 = add32+add31;
- add370 = add30+add29;
- add371 = add28+add27;
- add372 = add26+add25;
- add373 = add24+add23;
- add374 = add22+add21;
- add375 = add20+add19;
- add376 = add18+add17;
- add377 = add16+add15;
- add378 = add14+add13;
- add379 = add12+add11;
- add380 = add10+add9;
- add381 = add8+add7;
- add382 = add6+add5;
- add383 = add4+add3;
- add384 = add2+add1;
- add385 = add384+add383;
- add386 = add382+add381;
- add387 = add380+add379;
- add388 = add378+add377;
- add389 = add376+add375;
- add390 = add374+add373;
- add391 = add372+add371;
- add392 = add370+add369;
- add393 = add368+add367;
- add394 = add366+add365;
- add395 = add364+add363;
- add396 = add362+add361;
- add397 = add360+add359;
- add398 = add358+add357;
- add399 = add356+add355;
- add400 = add354+add353;
- add401 = add352+add351;
- add402 = add350+add349;
- add403 = add348+add347;
- add404 = add346+add345;
- add405 = add344+add343;
- add406 = add342+add341;
- add407 = add340+add339;
- add408 = add338+add337;
- add409 = add336+add335;
- add410 = add334+add333;
- add411 = add332+add331;
- add412 = add330+add329;
- add413 = add328+add327;
- add414 = add326+add325;
- add415 = add324+add323;
- add416 = add322+add321;
- add417 = add320+add319;
- add418 = add318+add317;
- add419 = add316+add315;
- add420 = add314+add313;
- add421 = add312+add311;
- add422 = add310+add309;
- add423 = add308+add307;
- add424 = add306+add305;
- add425 = add304+add303;
- add426 = add302+add301;
- add427 = add300+add299;
- add428 = add298+add297;
- add429 = add296+add295;
- add430 = add294+add293;
- add431 = add292+add291;
- add432 = add290+add289;
- add433 = add288+add287;
- add434 = add286+add285;
- add435 = add284+add283;
- add436 = add282+add281;
- add437 = add280+add279;
- add438 = add278+add277;
- add439 = add276+add275;
- add440 = add274+add273;
- add441 = add272+add271;
- add442 = add270+add269;
- add443 = add268+add267;
- add444 = add266+add265;
- add445 = add264+add263;
- add446 = add262+add261;
- add447 = add260+add259;
- add448 = add258+add257;
- add449 = add448+add447;
- add450 = add446+add445;
- add451 = add444+add443;
- add452 = add442+add441;
- add453 = add440+add439;
- add454 = add438+add437;
- add455 = add436+add435;
- add456 = add434+add433;
- add457 = add432+add431;
- add458 = add430+add429;
- add459 = add428+add427;
- add460 = add426+add425;
- add461 = add424+add423;
- add462 = add422+add421;
- add463 = add420+add419;
- add464 = add418+add417;
- add465 = add416+add415;
- add466 = add414+add413;
- add467 = add412+add411;
- add468 = add410+add409;
- add469 = add408+add407;
- add470 = add406+add405;
- add471 = add404+add403;
- add472 = add402+add401;
- add473 = add400+add399;
- add474 = add398+add397;
- add475 = add396+add395;
- add476 = add394+add393;
- add477 = add392+add391;
- add478 = add390+add389;
- add479 = add388+add387;
- add480 = add386+add385;
- add481 = add480+add479;
- add482 = add478+add477;
- add483 = add476+add475;
- add484 = add474+add473;
- add485 = add472+add471;
- add486 = add470+add469;
- add487 = add468+add467;
- add488 = add466+add465;
- add489 = add464+add463;
- add490 = add462+add461;
- add491 = add460+add459;
- add492 = add458+add457;
- add493 = add456+add455;
- add494 = add454+add453;
- add495 = add452+add451;
- add496 = add450+add449;
- add497 = add496+add495;
- add498 = add494+add493;
- add499 = add492+add491;
- add500 = add490+add489;
- add501 = add488+add487;
- add502 = add486+add485;
- add503 = add484+add483;
- add504 = add482+add481;
- add505 = add504+add503;
- add506 = add502+add501;
- add507 = add500+add499;
- add508 = add498+add497;
- add509 = add508+add507;
- add510 = add506+add505;
- return add509+add510;
-}
\ No newline at end of file
diff --git a/nmtdecoder/ce.h b/nmtdecoder/ce.h
deleted file mode 100644
index 62dda7b0a16a7d3ff6eaf5bf56eaaf2c415816b5..0000000000000000000000000000000000000000
--- a/nmtdecoder/ce.h
+++ /dev/null
@@ -1,1026 +0,0 @@
-#ifndef CE_H
-#define CE_H
-float compute_engine8(
-float a1, float b1,
-float a2, float b2,
-float a3, float b3,
-float a4, float b4,
-float a5, float b5,
-float a6, float b6,
-float a7, float b7,
-float a8, float b8);
-float compute_engine16(
-float a1, float b1,
-float a2, float b2,
-float a3, float b3,
-float a4, float b4,
-float a5, float b5,
-float a6, float b6,
-float a7, float b7,
-float a8, float b8,
-float a9, float b9,
-float a10, float b10,
-float a11, float b11,
-float a12, float b12,
-float a13, float b13,
-float a14, float b14,
-float a15, float b15,
-float a16, float b16);
-float compute_engine32(
-float a1, float b1,
-float a2, float b2,
-float a3, float b3,
-float a4, float b4,
-float a5, float b5,
-float a6, float b6,
-float a7, float b7,
-float a8, float b8,
-float a9, float b9,
-float a10, float b10,
-float a11, float b11,
-float a12, float b12,
-float a13, float b13,
-float a14, float b14,
-float a15, float b15,
-float a16, float b16,
-float a17, float b17,
-float a18, float b18,
-float a19, float b19,
-float a20, float b20,
-float a21, float b21,
-float a22, float b22,
-float a23, float b23,
-float a24, float b24,
-float a25, float b25,
-float a26, float b26,
-float a27, float b27,
-float a28, float b28,
-float a29, float b29,
-float a30, float b30,
-float a31, float b31,
-float a32, float b32);
-float compute_engine64(
-float a1, float b1,
-float a2, float b2,
-float a3, float b3,
-float a4, float b4,
-float a5, float b5,
-float a6, float b6,
-float a7, float b7,
-float a8, float b8,
-float a9, float b9,
-float a10, float b10,
-float a11, float b11,
-float a12, float b12,
-float a13, float b13,
-float a14, float b14,
-float a15, float b15,
-float a16, float b16,
-float a17, float b17,
-float a18, float b18,
-float a19, float b19,
-float a20, float b20,
-float a21, float b21,
-float a22, float b22,
-float a23, float b23,
-float a24, float b24,
-float a25, float b25,
-float a26, float b26,
-float a27, float b27,
-float a28, float b28,
-float a29, float b29,
-float a30, float b30,
-float a31, float b31,
-float a32, float b32,
-float a33, float b33,
-float a34, float b34,
-float a35, float b35,
-float a36, float b36,
-float a37, float b37,
-float a38, float b38,
-float a39, float b39,
-float a40, float b40,
-float a41, float b41,
-float a42, float b42,
-float a43, float b43,
-float a44, float b44,
-float a45, float b45,
-float a46, float b46,
-float a47, float b47,
-float a48, float b48,
-float a49, float b49,
-float a50, float b50,
-float a51, float b51,
-float a52, float b52,
-float a53, float b53,
-float a54, float b54,
-float a55, float b55,
-float a56, float b56,
-float a57, float b57,
-float a58, float b58,
-float a59, float b59,
-float a60, float b60,
-float a61, float b61,
-float a62, float b62,
-float a63, float b63,
-float a64, float b64);
-float compute_engine128(
-float a1, float b1,
-float a2, float b2,
-float a3, float b3,
-float a4, float b4,
-float a5, float b5,
-float a6, float b6,
-float a7, float b7,
-float a8, float b8,
-float a9, float b9,
-float a10, float b10,
-float a11, float b11,
-float a12, float b12,
-float a13, float b13,
-float a14, float b14,
-float a15, float b15,
-float a16, float b16,
-float a17, float b17,
-float a18, float b18,
-float a19, float b19,
-float a20, float b20,
-float a21, float b21,
-float a22, float b22,
-float a23, float b23,
-float a24, float b24,
-float a25, float b25,
-float a26, float b26,
-float a27, float b27,
-float a28, float b28,
-float a29, float b29,
-float a30, float b30,
-float a31, float b31,
-float a32, float b32,
-float a33, float b33,
-float a34, float b34,
-float a35, float b35,
-float a36, float b36,
-float a37, float b37,
-float a38, float b38,
-float a39, float b39,
-float a40, float b40,
-float a41, float b41,
-float a42, float b42,
-float a43, float b43,
-float a44, float b44,
-float a45, float b45,
-float a46, float b46,
-float a47, float b47,
-float a48, float b48,
-float a49, float b49,
-float a50, float b50,
-float a51, float b51,
-float a52, float b52,
-float a53, float b53,
-float a54, float b54,
-float a55, float b55,
-float a56, float b56,
-float a57, float b57,
-float a58, float b58,
-float a59, float b59,
-float a60, float b60,
-float a61, float b61,
-float a62, float b62,
-float a63, float b63,
-float a64, float b64,
-float a65, float b65,
-float a66, float b66,
-float a67, float b67,
-float a68, float b68,
-float a69, float b69,
-float a70, float b70,
-float a71, float b71,
-float a72, float b72,
-float a73, float b73,
-float a74, float b74,
-float a75, float b75,
-float a76, float b76,
-float a77, float b77,
-float a78, float b78,
-float a79, float b79,
-float a80, float b80,
-float a81, float b81,
-float a82, float b82,
-float a83, float b83,
-float a84, float b84,
-float a85, float b85,
-float a86, float b86,
-float a87, float b87,
-float a88, float b88,
-float a89, float b89,
-float a90, float b90,
-float a91, float b91,
-float a92, float b92,
-float a93, float b93,
-float a94, float b94,
-float a95, float b95,
-float a96, float b96,
-float a97, float b97,
-float a98, float b98,
-float a99, float b99,
-float a100, float b100,
-float a101, float b101,
-float a102, float b102,
-float a103, float b103,
-float a104, float b104,
-float a105, float b105,
-float a106, float b106,
-float a107, float b107,
-float a108, float b108,
-float a109, float b109,
-float a110, float b110,
-float a111, float b111,
-float a112, float b112,
-float a113, float b113,
-float a114, float b114,
-float a115, float b115,
-float a116, float b116,
-float a117, float b117,
-float a118, float b118,
-float a119, float b119,
-float a120, float b120,
-float a121, float b121,
-float a122, float b122,
-float a123, float b123,
-float a124, float b124,
-float a125, float b125,
-float a126, float b126,
-float a127, float b127,
-float a128, float b128);
-float compute_engine256(
-float a1, float b1,
-float a2, float b2,
-float a3, float b3,
-float a4, float b4,
-float a5, float b5,
-float a6, float b6,
-float a7, float b7,
-float a8, float b8,
-float a9, float b9,
-float a10, float b10,
-float a11, float b11,
-float a12, float b12,
-float a13, float b13,
-float a14, float b14,
-float a15, float b15,
-float a16, float b16,
-float a17, float b17,
-float a18, float b18,
-float a19, float b19,
-float a20, float b20,
-float a21, float b21,
-float a22, float b22,
-float a23, float b23,
-float a24, float b24,
-float a25, float b25,
-float a26, float b26,
-float a27, float b27,
-float a28, float b28,
-float a29, float b29,
-float a30, float b30,
-float a31, float b31,
-float a32, float b32,
-float a33, float b33,
-float a34, float b34,
-float a35, float b35,
-float a36, float b36,
-float a37, float b37,
-float a38, float b38,
-float a39, float b39,
-float a40, float b40,
-float a41, float b41,
-float a42, float b42,
-float a43, float b43,
-float a44, float b44,
-float a45, float b45,
-float a46, float b46,
-float a47, float b47,
-float a48, float b48,
-float a49, float b49,
-float a50, float b50,
-float a51, float b51,
-float a52, float b52,
-float a53, float b53,
-float a54, float b54,
-float a55, float b55,
-float a56, float b56,
-float a57, float b57,
-float a58, float b58,
-float a59, float b59,
-float a60, float b60,
-float a61, float b61,
-float a62, float b62,
-float a63, float b63,
-float a64, float b64,
-float a65, float b65,
-float a66, float b66,
-float a67, float b67,
-float a68, float b68,
-float a69, float b69,
-float a70, float b70,
-float a71, float b71,
-float a72, float b72,
-float a73, float b73,
-float a74, float b74,
-float a75, float b75,
-float a76, float b76,
-float a77, float b77,
-float a78, float b78,
-float a79, float b79,
-float a80, float b80,
-float a81, float b81,
-float a82, float b82,
-float a83, float b83,
-float a84, float b84,
-float a85, float b85,
-float a86, float b86,
-float a87, float b87,
-float a88, float b88,
-float a89, float b89,
-float a90, float b90,
-float a91, float b91,
-float a92, float b92,
-float a93, float b93,
-float a94, float b94,
-float a95, float b95,
-float a96, float b96,
-float a97, float b97,
-float a98, float b98,
-float a99, float b99,
-float a100, float b100,
-float a101, float b101,
-float a102, float b102,
-float a103, float b103,
-float a104, float b104,
-float a105, float b105,
-float a106, float b106,
-float a107, float b107,
-float a108, float b108,
-float a109, float b109,
-float a110, float b110,
-float a111, float b111,
-float a112, float b112,
-float a113, float b113,
-float a114, float b114,
-float a115, float b115,
-float a116, float b116,
-float a117, float b117,
-float a118, float b118,
-float a119, float b119,
-float a120, float b120,
-float a121, float b121,
-float a122, float b122,
-float a123, float b123,
-float a124, float b124,
-float a125, float b125,
-float a126, float b126,
-float a127, float b127,
-float a128, float b128,
-float a129, float b129,
-float a130, float b130,
-float a131, float b131,
-float a132, float b132,
-float a133, float b133,
-float a134, float b134,
-float a135, float b135,
-float a136, float b136,
-float a137, float b137,
-float a138, float b138,
-float a139, float b139,
-float a140, float b140,
-float a141, float b141,
-float a142, float b142,
-float a143, float b143,
-float a144, float b144,
-float a145, float b145,
-float a146, float b146,
-float a147, float b147,
-float a148, float b148,
-float a149, float b149,
-float a150, float b150,
-float a151, float b151,
-float a152, float b152,
-float a153, float b153,
-float a154, float b154,
-float a155, float b155,
-float a156, float b156,
-float a157, float b157,
-float a158, float b158,
-float a159, float b159,
-float a160, float b160,
-float a161, float b161,
-float a162, float b162,
-float a163, float b163,
-float a164, float b164,
-float a165, float b165,
-float a166, float b166,
-float a167, float b167,
-float a168, float b168,
-float a169, float b169,
-float a170, float b170,
-float a171, float b171,
-float a172, float b172,
-float a173, float b173,
-float a174, float b174,
-float a175, float b175,
-float a176, float b176,
-float a177, float b177,
-float a178, float b178,
-float a179, float b179,
-float a180, float b180,
-float a181, float b181,
-float a182, float b182,
-float a183, float b183,
-float a184, float b184,
-float a185, float b185,
-float a186, float b186,
-float a187, float b187,
-float a188, float b188,
-float a189, float b189,
-float a190, float b190,
-float a191, float b191,
-float a192, float b192,
-float a193, float b193,
-float a194, float b194,
-float a195, float b195,
-float a196, float b196,
-float a197, float b197,
-float a198, float b198,
-float a199, float b199,
-float a200, float b200,
-float a201, float b201,
-float a202, float b202,
-float a203, float b203,
-float a204, float b204,
-float a205, float b205,
-float a206, float b206,
-float a207, float b207,
-float a208, float b208,
-float a209, float b209,
-float a210, float b210,
-float a211, float b211,
-float a212, float b212,
-float a213, float b213,
-float a214, float b214,
-float a215, float b215,
-float a216, float b216,
-float a217, float b217,
-float a218, float b218,
-float a219, float b219,
-float a220, float b220,
-float a221, float b221,
-float a222, float b222,
-float a223, float b223,
-float a224, float b224,
-float a225, float b225,
-float a226, float b226,
-float a227, float b227,
-float a228, float b228,
-float a229, float b229,
-float a230, float b230,
-float a231, float b231,
-float a232, float b232,
-float a233, float b233,
-float a234, float b234,
-float a235, float b235,
-float a236, float b236,
-float a237, float b237,
-float a238, float b238,
-float a239, float b239,
-float a240, float b240,
-float a241, float b241,
-float a242, float b242,
-float a243, float b243,
-float a244, float b244,
-float a245, float b245,
-float a246, float b246,
-float a247, float b247,
-float a248, float b248,
-float a249, float b249,
-float a250, float b250,
-float a251, float b251,
-float a252, float b252,
-float a253, float b253,
-float a254, float b254,
-float a255, float b255,
-float a256, float b256);
-float compute_engine512(
-float a1, float b1,
-float a2, float b2,
-float a3, float b3,
-float a4, float b4,
-float a5, float b5,
-float a6, float b6,
-float a7, float b7,
-float a8, float b8,
-float a9, float b9,
-float a10, float b10,
-float a11, float b11,
-float a12, float b12,
-float a13, float b13,
-float a14, float b14,
-float a15, float b15,
-float a16, float b16,
-float a17, float b17,
-float a18, float b18,
-float a19, float b19,
-float a20, float b20,
-float a21, float b21,
-float a22, float b22,
-float a23, float b23,
-float a24, float b24,
-float a25, float b25,
-float a26, float b26,
-float a27, float b27,
-float a28, float b28,
-float a29, float b29,
-float a30, float b30,
-float a31, float b31,
-float a32, float b32,
-float a33, float b33,
-float a34, float b34,
-float a35, float b35,
-float a36, float b36,
-float a37, float b37,
-float a38, float b38,
-float a39, float b39,
-float a40, float b40,
-float a41, float b41,
-float a42, float b42,
-float a43, float b43,
-float a44, float b44,
-float a45, float b45,
-float a46, float b46,
-float a47, float b47,
-float a48, float b48,
-float a49, float b49,
-float a50, float b50,
-float a51, float b51,
-float a52, float b52,
-float a53, float b53,
-float a54, float b54,
-float a55, float b55,
-float a56, float b56,
-float a57, float b57,
-float a58, float b58,
-float a59, float b59,
-float a60, float b60,
-float a61, float b61,
-float a62, float b62,
-float a63, float b63,
-float a64, float b64,
-float a65, float b65,
-float a66, float b66,
-float a67, float b67,
-float a68, float b68,
-float a69, float b69,
-float a70, float b70,
-float a71, float b71,
-float a72, float b72,
-float a73, float b73,
-float a74, float b74,
-float a75, float b75,
-float a76, float b76,
-float a77, float b77,
-float a78, float b78,
-float a79, float b79,
-float a80, float b80,
-float a81, float b81,
-float a82, float b82,
-float a83, float b83,
-float a84, float b84,
-float a85, float b85,
-float a86, float b86,
-float a87, float b87,
-float a88, float b88,
-float a89, float b89,
-float a90, float b90,
-float a91, float b91,
-float a92, float b92,
-float a93, float b93,
-float a94, float b94,
-float a95, float b95,
-float a96, float b96,
-float a97, float b97,
-float a98, float b98,
-float a99, float b99,
-float a100, float b100,
-float a101, float b101,
-float a102, float b102,
-float a103, float b103,
-float a104, float b104,
-float a105, float b105,
-float a106, float b106,
-float a107, float b107,
-float a108, float b108,
-float a109, float b109,
-float a110, float b110,
-float a111, float b111,
-float a112, float b112,
-float a113, float b113,
-float a114, float b114,
-float a115, float b115,
-float a116, float b116,
-float a117, float b117,
-float a118, float b118,
-float a119, float b119,
-float a120, float b120,
-float a121, float b121,
-float a122, float b122,
-float a123, float b123,
-float a124, float b124,
-float a125, float b125,
-float a126, float b126,
-float a127, float b127,
-float a128, float b128,
-float a129, float b129,
-float a130, float b130,
-float a131, float b131,
-float a132, float b132,
-float a133, float b133,
-float a134, float b134,
-float a135, float b135,
-float a136, float b136,
-float a137, float b137,
-float a138, float b138,
-float a139, float b139,
-float a140, float b140,
-float a141, float b141,
-float a142, float b142,
-float a143, float b143,
-float a144, float b144,
-float a145, float b145,
-float a146, float b146,
-float a147, float b147,
-float a148, float b148,
-float a149, float b149,
-float a150, float b150,
-float a151, float b151,
-float a152, float b152,
-float a153, float b153,
-float a154, float b154,
-float a155, float b155,
-float a156, float b156,
-float a157, float b157,
-float a158, float b158,
-float a159, float b159,
-float a160, float b160,
-float a161, float b161,
-float a162, float b162,
-float a163, float b163,
-float a164, float b164,
-float a165, float b165,
-float a166, float b166,
-float a167, float b167,
-float a168, float b168,
-float a169, float b169,
-float a170, float b170,
-float a171, float b171,
-float a172, float b172,
-float a173, float b173,
-float a174, float b174,
-float a175, float b175,
-float a176, float b176,
-float a177, float b177,
-float a178, float b178,
-float a179, float b179,
-float a180, float b180,
-float a181, float b181,
-float a182, float b182,
-float a183, float b183,
-float a184, float b184,
-float a185, float b185,
-float a186, float b186,
-float a187, float b187,
-float a188, float b188,
-float a189, float b189,
-float a190, float b190,
-float a191, float b191,
-float a192, float b192,
-float a193, float b193,
-float a194, float b194,
-float a195, float b195,
-float a196, float b196,
-float a197, float b197,
-float a198, float b198,
-float a199, float b199,
-float a200, float b200,
-float a201, float b201,
-float a202, float b202,
-float a203, float b203,
-float a204, float b204,
-float a205, float b205,
-float a206, float b206,
-float a207, float b207,
-float a208, float b208,
-float a209, float b209,
-float a210, float b210,
-float a211, float b211,
-float a212, float b212,
-float a213, float b213,
-float a214, float b214,
-float a215, float b215,
-float a216, float b216,
-float a217, float b217,
-float a218, float b218,
-float a219, float b219,
-float a220, float b220,
-float a221, float b221,
-float a222, float b222,
-float a223, float b223,
-float a224, float b224,
-float a225, float b225,
-float a226, float b226,
-float a227, float b227,
-float a228, float b228,
-float a229, float b229,
-float a230, float b230,
-float a231, float b231,
-float a232, float b232,
-float a233, float b233,
-float a234, float b234,
-float a235, float b235,
-float a236, float b236,
-float a237, float b237,
-float a238, float b238,
-float a239, float b239,
-float a240, float b240,
-float a241, float b241,
-float a242, float b242,
-float a243, float b243,
-float a244, float b244,
-float a245, float b245,
-float a246, float b246,
-float a247, float b247,
-float a248, float b248,
-float a249, float b249,
-float a250, float b250,
-float a251, float b251,
-float a252, float b252,
-float a253, float b253,
-float a254, float b254,
-float a255, float b255,
-float a256, float b256,
-float a257, float b257,
-float a258, float b258,
-float a259, float b259,
-float a260, float b260,
-float a261, float b261,
-float a262, float b262,
-float a263, float b263,
-float a264, float b264,
-float a265, float b265,
-float a266, float b266,
-float a267, float b267,
-float a268, float b268,
-float a269, float b269,
-float a270, float b270,
-float a271, float b271,
-float a272, float b272,
-float a273, float b273,
-float a274, float b274,
-float a275, float b275,
-float a276, float b276,
-float a277, float b277,
-float a278, float b278,
-float a279, float b279,
-float a280, float b280,
-float a281, float b281,
-float a282, float b282,
-float a283, float b283,
-float a284, float b284,
-float a285, float b285,
-float a286, float b286,
-float a287, float b287,
-float a288, float b288,
-float a289, float b289,
-float a290, float b290,
-float a291, float b291,
-float a292, float b292,
-float a293, float b293,
-float a294, float b294,
-float a295, float b295,
-float a296, float b296,
-float a297, float b297,
-float a298, float b298,
-float a299, float b299,
-float a300, float b300,
-float a301, float b301,
-float a302, float b302,
-float a303, float b303,
-float a304, float b304,
-float a305, float b305,
-float a306, float b306,
-float a307, float b307,
-float a308, float b308,
-float a309, float b309,
-float a310, float b310,
-float a311, float b311,
-float a312, float b312,
-float a313, float b313,
-float a314, float b314,
-float a315, float b315,
-float a316, float b316,
-float a317, float b317,
-float a318, float b318,
-float a319, float b319,
-float a320, float b320,
-float a321, float b321,
-float a322, float b322,
-float a323, float b323,
-float a324, float b324,
-float a325, float b325,
-float a326, float b326,
-float a327, float b327,
-float a328, float b328,
-float a329, float b329,
-float a330, float b330,
-float a331, float b331,
-float a332, float b332,
-float a333, float b333,
-float a334, float b334,
-float a335, float b335,
-float a336, float b336,
-float a337, float b337,
-float a338, float b338,
-float a339, float b339,
-float a340, float b340,
-float a341, float b341,
-float a342, float b342,
-float a343, float b343,
-float a344, float b344,
-float a345, float b345,
-float a346, float b346,
-float a347, float b347,
-float a348, float b348,
-float a349, float b349,
-float a350, float b350,
-float a351, float b351,
-float a352, float b352,
-float a353, float b353,
-float a354, float b354,
-float a355, float b355,
-float a356, float b356,
-float a357, float b357,
-float a358, float b358,
-float a359, float b359,
-float a360, float b360,
-float a361, float b361,
-float a362, float b362,
-float a363, float b363,
-float a364, float b364,
-float a365, float b365,
-float a366, float b366,
-float a367, float b367,
-float a368, float b368,
-float a369, float b369,
-float a370, float b370,
-float a371, float b371,
-float a372, float b372,
-float a373, float b373,
-float a374, float b374,
-float a375, float b375,
-float a376, float b376,
-float a377, float b377,
-float a378, float b378,
-float a379, float b379,
-float a380, float b380,
-float a381, float b381,
-float a382, float b382,
-float a383, float b383,
-float a384, float b384,
-float a385, float b385,
-float a386, float b386,
-float a387, float b387,
-float a388, float b388,
-float a389, float b389,
-float a390, float b390,
-float a391, float b391,
-float a392, float b392,
-float a393, float b393,
-float a394, float b394,
-float a395, float b395,
-float a396, float b396,
-float a397, float b397,
-float a398, float b398,
-float a399, float b399,
-float a400, float b400,
-float a401, float b401,
-float a402, float b402,
-float a403, float b403,
-float a404, float b404,
-float a405, float b405,
-float a406, float b406,
-float a407, float b407,
-float a408, float b408,
-float a409, float b409,
-float a410, float b410,
-float a411, float b411,
-float a412, float b412,
-float a413, float b413,
-float a414, float b414,
-float a415, float b415,
-float a416, float b416,
-float a417, float b417,
-float a418, float b418,
-float a419, float b419,
-float a420, float b420,
-float a421, float b421,
-float a422, float b422,
-float a423, float b423,
-float a424, float b424,
-float a425, float b425,
-float a426, float b426,
-float a427, float b427,
-float a428, float b428,
-float a429, float b429,
-float a430, float b430,
-float a431, float b431,
-float a432, float b432,
-float a433, float b433,
-float a434, float b434,
-float a435, float b435,
-float a436, float b436,
-float a437, float b437,
-float a438, float b438,
-float a439, float b439,
-float a440, float b440,
-float a441, float b441,
-float a442, float b442,
-float a443, float b443,
-float a444, float b444,
-float a445, float b445,
-float a446, float b446,
-float a447, float b447,
-float a448, float b448,
-float a449, float b449,
-float a450, float b450,
-float a451, float b451,
-float a452, float b452,
-float a453, float b453,
-float a454, float b454,
-float a455, float b455,
-float a456, float b456,
-float a457, float b457,
-float a458, float b458,
-float a459, float b459,
-float a460, float b460,
-float a461, float b461,
-float a462, float b462,
-float a463, float b463,
-float a464, float b464,
-float a465, float b465,
-float a466, float b466,
-float a467, float b467,
-float a468, float b468,
-float a469, float b469,
-float a470, float b470,
-float a471, float b471,
-float a472, float b472,
-float a473, float b473,
-float a474, float b474,
-float a475, float b475,
-float a476, float b476,
-float a477, float b477,
-float a478, float b478,
-float a479, float b479,
-float a480, float b480,
-float a481, float b481,
-float a482, float b482,
-float a483, float b483,
-float a484, float b484,
-float a485, float b485,
-float a486, float b486,
-float a487, float b487,
-float a488, float b488,
-float a489, float b489,
-float a490, float b490,
-float a491, float b491,
-float a492, float b492,
-float a493, float b493,
-float a494, float b494,
-float a495, float b495,
-float a496, float b496,
-float a497, float b497,
-float a498, float b498,
-float a499, float b499,
-float a500, float b500,
-float a501, float b501,
-float a502, float b502,
-float a503, float b503,
-float a504, float b504,
-float a505, float b505,
-float a506, float b506,
-float a507, float b507,
-float a508, float b508,
-float a509, float b509,
-float a510, float b510,
-float a511, float b511,
-float a512, float b512);
-#endif
\ No newline at end of file
diff --git a/nmtdecoder/layers.cc b/nmtdecoder/layers.cc
deleted file mode 100644
index f2b1dd6e528b2a67461e52b7492892248cf84d1c..0000000000000000000000000000000000000000
--- a/nmtdecoder/layers.cc
+++ /dev/null
@@ -1,330 +0,0 @@
-#include "layers.h"
-//#include <stdio.h>
-#include <math.h>
-float exp_hls(float in){
-/*
-union
-{
-unsigned int i;
-float f;
-}v;
-v.i=(1<<23)*(1.4426950409*in+126.94201519f);
-return v.f;
-*/
-return expf(in);
-}
-
-float sigmoid(float x) {
- return 1. / (1. + exp_hls(-x));
-}
-
-float tanh_hls(float x){
- //return (exp_hls(2*x)-1)/(exp_hls(2*x)+1);
- return tanhf(x);
-}
-float z[dim];//update gate
- float r[dim];//reset gate
- float h_[dim];//new memory
-
-
-/*memory is for the last state*/
-void gru_layer_base(ap_int<512> *data,float input[dim_word], float output[dim],float memory[dim],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
- #pragma HLS INTERFACE m_axi port=data
- int i,kk;
- /*float z[dim];//update gate
- float r[dim];//reset gate
- float h_[dim];//new memory*/
- float bx[dim];
- int index=0;
-
- #pragma HLS ARRAY_PARTITION variable=bx cyclic dim=1 factor = 16
- for(i = 0;i < dim; i++){
- h_[i]=0.0;
- }
-
- for(i = 0;i < dim; i+=16){
- for (kk=0; kk<16; kk++){
- r[i+kk]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(i = 0;i < dim; i+=16){
- for (kk=0; kk<16; kk++){
- z[i+kk]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
- mvm512_1024_2out(data+128, input, r, z,kvector,kmatrix,kout);
- //for(i = 0;i < dim; i++)
- //printf("r %f\n",r[i]);
- //for(i = 0;i < dim; i++)
- //printf("memory %f\n",memory[i]);
- mvm1024_1024_2out(data+65664, memory,r, z,kvector,kmatrix,kout);
- //for(i = 0;i < dim; i++)
- //printf("r %f\n",r[i]);
-
- for(i = 0;i < dim; i++){
- r[i]=sigmoid(r[i]);
- z[i]=sigmoid(z[i]);
-
- }
-
- mvm1024_1024(data+196736, memory,h_,kvector,kmatrix,kout);
- //for(i = 0;i < dim; i++)
- //printf("h1 %f\n",h_[i]);
- for(i = 0;i < dim; i++)
- h_[i]=r[i]*h_[i];
- mvm512_1024(data+262272, input,h_,kvector,kmatrix,kout);
- //for(i = 0;i < dim; i++)
- //printf("h %f\n",h_[i]);
- index=295040;
- for(i = 0;i < dim; i+=16){
- for (kk=0; kk<16; kk++){
- bx[i+kk]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
-
- for(i = 0;i < dim; i++){
- h_[i] = tanh_hls(h_[i] + bx[i]);
- output[i] =(1.0 - z[i])* h_[i] + z[i] * memory[i];
- }
-
-
-}
-
-
-void gru_layer_nl(ap_int<512> *data,float input[ctxdim], float output[dim],float memory[dim],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
- #pragma HLS INTERFACE m_axi port=data
- int i,kk,index;
- /*float z[dim];//update gate
- float r[dim];//reset gate
- float h_[dim];//new memory*/
-
- index=0;
- for(i = 0;i < dim; i+=16){
- for (kk=0; kk<16; kk++){
- r[i+kk]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(i = 0;i < dim; i+=16){
- for (kk=0; kk<16; kk++){
- z[i+kk]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(i = 0;i < dim; i+=16){
- for (kk=0; kk<16; kk++){
- h_[i+kk]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- mvm2048_1024_2out(data+192, input, r, z,kvector,kmatrix,kout);
- mvm1024_1024_2out(data+262336, memory,r, z,kvector,kmatrix,kout);
- for(i = 0;i < dim; i++){
- r[i]=sigmoid(r[i]);
- z[i]=sigmoid(z[i]);
-
- }
-
- mvm1024_1024(data+393408, memory,h_,kvector,kmatrix,kout);
- for(i = 0;i < dim; i++)
- h_[i]=r[i]*(h_[i]);
- mvm2048_1024(data+458944, input,h_,kvector,kmatrix,kout);
- for(i = 0;i < dim; i++){
-//#pragma HLS unroll factor = 16
- h_[i] = tanh_hls(h_[i]);
- output[i] =(1.0 - z[i])* h_[i] + z[i] * memory[i];
- }
-}
-
-void attention(ap_int<512> *data,float input_h1[dim], float input_f[MAXSTEP][dim], float input_b[MAXSTEP][dim], int stepsize, float output[ctxdim],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
-//#pragma HLS inline off
-//#pragma HLS inline region recursive
-//#pragma HLS ALLOCATION instances=compute_engine64 limit=1 function
-#pragma HLS INTERFACE m_axi port=data
- int i,j,k,kk,index;
- index=0;
- float s;
- float alpha[MAXSTEP];
- float temp[ctxdim];
- float decoder_b_att[ctxdim];
- float decoder_U_att[ctxdim];
- float decoder_c_tt;
-
-#pragma HLS ARRAY_PARTITION variable=decoder_U_att cyclic dim=1 factor = 16
-#pragma HLS ARRAY_PARTITION variable=decoder_b_att cyclic dim=1 factor = 16
- for(j=0; j< ctxdim; j++){
- output[j]=0.0; // clear output
-
- }
-
- for(i = 0;i < ctxdim; i+=16){
- for (kk=0; kk<16; kk++){
- decoder_b_att[i+kk]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(i = 0;i < ctxdim; i+=16){
- for (kk=0; kk<16; kk++){
- decoder_U_att[i+kk]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- decoder_c_tt=trans_i(data[256].range(31,0));
- for(i=0; i< MAXSTEP; i++){
- if(i<stepsize){
- for(j=0; j< ctxdim; j++)
- temp[j]=decoder_b_att[j];
- mvm1024_2048(data+257, input_h1,temp,kvector,kmatrix,kout);
- mvm1024_2048_2in(data+131329,input_f[i],input_b[stepsize-i-1],temp,kvector,kmatrix,kout);
- for(j=0; j< ctxdim; j++){
-//#pragma HLS unroll factor = 8
- temp[j]=tanh_hls(temp[j]);
- }
- alpha[i] = 0.0;
- for(j=0; j< ctxdim; j+=64){
-#pragma HLS pipeline
- alpha[i] += compute_engine64(temp[j+0],decoder_U_att[j+0],
- temp[j+1],decoder_U_att[j+1],
- temp[j+2],decoder_U_att[j+2],
- temp[j+3],decoder_U_att[j+3],
- temp[j+4],decoder_U_att[j+4],
- temp[j+5],decoder_U_att[j+5],
- temp[j+6],decoder_U_att[j+6],
- temp[j+7],decoder_U_att[j+7],
- temp[j+8],decoder_U_att[j+8],
- temp[j+9],decoder_U_att[j+9],
- temp[j+10],decoder_U_att[j+10],
- temp[j+11],decoder_U_att[j+11],
- temp[j+12],decoder_U_att[j+12],
- temp[j+13],decoder_U_att[j+13],
- temp[j+14],decoder_U_att[j+14],
- temp[j+15],decoder_U_att[j+15],
- temp[j+16],decoder_U_att[j+16],
- temp[j+17],decoder_U_att[j+17],
- temp[j+18],decoder_U_att[j+18],
- temp[j+19],decoder_U_att[j+19],
- temp[j+20],decoder_U_att[j+20],
- temp[j+21],decoder_U_att[j+21],
- temp[j+22],decoder_U_att[j+22],
- temp[j+23],decoder_U_att[j+23],
- temp[j+24],decoder_U_att[j+24],
- temp[j+25],decoder_U_att[j+25],
- temp[j+26],decoder_U_att[j+26],
- temp[j+27],decoder_U_att[j+27],
- temp[j+28],decoder_U_att[j+28],
- temp[j+29],decoder_U_att[j+29],
- temp[j+30],decoder_U_att[j+30],
- temp[j+31],decoder_U_att[j+31],
- temp[j+32],decoder_U_att[j+32],
- temp[j+33],decoder_U_att[j+33],
- temp[j+34],decoder_U_att[j+34],
- temp[j+35],decoder_U_att[j+35],
- temp[j+36],decoder_U_att[j+36],
- temp[j+37],decoder_U_att[j+37],
- temp[j+38],decoder_U_att[j+38],
- temp[j+39],decoder_U_att[j+39],
- temp[j+40],decoder_U_att[j+40],
- temp[j+41],decoder_U_att[j+41],
- temp[j+42],decoder_U_att[j+42],
- temp[j+43],decoder_U_att[j+43],
- temp[j+44],decoder_U_att[j+44],
- temp[j+45],decoder_U_att[j+45],
- temp[j+46],decoder_U_att[j+46],
- temp[j+47],decoder_U_att[j+47],
- temp[j+48],decoder_U_att[j+48],
- temp[j+49],decoder_U_att[j+49],
- temp[j+50],decoder_U_att[j+50],
- temp[j+51],decoder_U_att[j+51],
- temp[j+52],decoder_U_att[j+52],
- temp[j+53],decoder_U_att[j+53],
- temp[j+54],decoder_U_att[j+54],
- temp[j+55],decoder_U_att[j+55],
- temp[j+56],decoder_U_att[j+56],
- temp[j+57],decoder_U_att[j+57],
- temp[j+58],decoder_U_att[j+58],
- temp[j+59],decoder_U_att[j+59],
- temp[j+60],decoder_U_att[j+60],
- temp[j+61],decoder_U_att[j+61],
- temp[j+62],decoder_U_att[j+62],
- temp[j+63],decoder_U_att[j+63]);
- }
- alpha[i]= exp_hls(alpha[i] + decoder_c_tt); //alpha = exp(alpha)
- }
- }
- s = 0.0;
- for(i=0; i< MAXSTEP; i++){
- if(i<stepsize){
- s += alpha[i];
- }
- }
- for(i=0; i< MAXSTEP; i++){ // ctx = sum(alpha * hi)
- if(i<stepsize){
- alpha[i] = alpha[i]/s;
- //printf("alpha: %f \n", alpha[i]);
- for(j =0; j < dim; j++){
- output[j] += alpha[i]* input_f[i][j];
- output[j+dim] += alpha[i]* input_b[stepsize-i-1][j];
- }
- }
- }
-}
-void logit_lstm(ap_int<512> *data,float input[dim], float output[dim_word],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
- #pragma HLS INTERFACE m_axi port=data
- int i,kk;
- int index=0;
- for(i = 0;i < dim_word; i+=16){
- for (kk=0; kk<16; kk++){
- output[i+kk]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- mvm1024_512(data+32, input,output,kvector,kmatrix,kout);
-
-}
-
-void logit_prev(ap_int<512> *data,float input[dim_word], float output[dim_word],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
- #pragma HLS INTERFACE m_axi port=data
- int i,kk;
- int index=0;
- for(i = 0;i < dim_word; i+=16){
- for (kk=0; kk<16; kk++){
- output[i+kk]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- mvm512_512(data+32, input,output,kvector,kmatrix,kout);
-
-}
-
-void logit_ctx(ap_int<512> *data,float input[ctxdim], float output[dim_word],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
- #pragma HLS INTERFACE m_axi port=data
- int i,kk;
- int index=0;
- for(i = 0;i < dim_word; i+=16){
- for (kk=0; kk<16; kk++){
- output[i+kk]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- mvm2048_512(data+32, input,output,kvector,kmatrix,kout);
-}
-
-void logit(ap_int<512> *data,float input[dim_word], float output[n_words],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
- #pragma HLS INTERFACE m_axi port=data
- int i,kk;
- int index=0;
- for(i = 0;i < n_words; i+=16){
- for (kk=0; kk<16; kk++){
- output[i+kk]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
- mvm512_30000(data+1875, input,output,kvector,kmatrix,kout);
-}
-
diff --git a/nmtdecoder/layers.h b/nmtdecoder/layers.h
deleted file mode 100644
index 7e3b76d0bb60954fe4f8abd36d85c4ce7206c97a..0000000000000000000000000000000000000000
--- a/nmtdecoder/layers.h
+++ /dev/null
@@ -1,64 +0,0 @@
-#ifndef LAYERS_H
-#define LAYERS_H
-#include "mvm.h"
-#define dim_word 512 // word vector dimensionality
-#define dim 1024 //the number of LSTM units
-#define n_words_src 30000 // source vocabulary size
-#define n_words 30000 // target vocabulary size
-#define ctxdim 2*dim //context dimension
-
-#define MAXSTEP 50 //maximum encoder step
-#define MAXSTEP_DE 50 // maximum decoder step
-
-#define MAX_WORD 1000
-#define MAX_SEN 100
-
-#define K 5 //beam size
-
-float exp_hls(float in);
-
-/*sigmoid function*/
-float sigmoid(float x);
-/*tanh function*/
-float tanh_hls(float x);
-
-/*
-input: emb (n_timesteps, n_samples, dim_word)
-output: h dim
-*/
-
-void gru_layer_base(ap_int<512> *data,float input[dim_word], float output[dim],float memory[dim],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-
-/*
-input: ctxdim
-output:dim
-*/
-void gru_layer_nl(ap_int<512> *data,float input[ctxdim], float output[dim],float memory[dim],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-
-/*
-input: all of the context
-output:ctxdim
-*/
-void attention(ap_int<512> *data,float input_h1[dim],float input_f[MAXSTEP][dim], float input_b[MAXSTEP][dim], int stepsize, float output[ctxdim],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-/*
-input:next_state - hidden states of the decoder gru
-output:dim_word
-*/
-void logit_lstm(ap_int<512> *data,float input[dim], float output[dim_word],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-/*
-input: emb
-output:dim_word
-*/
-void logit_prev(ap_int<512> *data,float input[dim_word], float output[dim_word],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-/*
-input:ctx weighted averages of context, generated by attention module
-output:dim_word
-*/
-void logit_ctx(ap_int<512> *data,float input[ctxdim], float output[dim_word],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-/*
-input:dim_word
-output:n_words
-*/
-void logit(ap_int<512> *data,float input[dim_word],float output[n_words],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-
-#endif
diff --git a/nmtdecoder/mvm.cc b/nmtdecoder/mvm.cc
deleted file mode 100644
index 97205970755def10c69a2c94f14d4f75d6c01ae7..0000000000000000000000000000000000000000
--- a/nmtdecoder/mvm.cc
+++ /dev/null
@@ -1,833 +0,0 @@
-#include "mvm.h"
-#include <math.h>
-//#include<stdio.h>
-
-float trans_i(ap_int<32> a){
- u temp;
- temp.i=a;
- return temp.f;
-
-}
-void kernel(float w[K_SIZE][K_NUM], float x[K_SIZE],float out[K_NUM]){
-#pragma HLS ARRAY_PARTITION variable=w cyclic dim=1 factor = 64
-#pragma HLS ARRAY_PARTITION variable=w cyclic dim=2 factor = 2
-#pragma HLS ARRAY_PARTITION variable=x cyclic dim=1 factor = 64
-#pragma HLS ARRAY_PARTITION variable=out cyclic dim=1 factor = 2
-//#pragma HLS ALLOCATION instances=compute_engine64 limit=2 function
-
- out[0]=compute_engine64(x[0],w[0][0],
- x[1],w[1][0],
- x[2],w[2][0],
- x[3],w[3][0],
- x[4],w[4][0],
- x[5],w[5][0],
- x[6],w[6][0],
- x[7],w[7][0],
- x[8],w[8][0],
- x[9],w[9][0],
- x[10],w[10][0],
- x[11],w[11][0],
- x[12],w[12][0],
- x[13],w[13][0],
- x[14],w[14][0],
- x[15],w[15][0],
- x[16],w[16][0],
- x[17],w[17][0],
- x[18],w[18][0],
- x[19],w[19][0],
- x[20],w[20][0],
- x[21],w[21][0],
- x[22],w[22][0],
- x[23],w[23][0],
- x[24],w[24][0],
- x[25],w[25][0],
- x[26],w[26][0],
- x[27],w[27][0],
- x[28],w[28][0],
- x[29],w[29][0],
- x[30],w[30][0],
- x[31],w[31][0],
- x[32],w[32][0],
- x[33],w[33][0],
- x[34],w[34][0],
- x[35],w[35][0],
- x[36],w[36][0],
- x[37],w[37][0],
- x[38],w[38][0],
- x[39],w[39][0],
- x[40],w[40][0],
- x[41],w[41][0],
- x[42],w[42][0],
- x[43],w[43][0],
- x[44],w[44][0],
- x[45],w[45][0],
- x[46],w[46][0],
- x[47],w[47][0],
- x[48],w[48][0],
- x[49],w[49][0],
- x[50],w[50][0],
- x[51],w[51][0],
- x[52],w[52][0],
- x[53],w[53][0],
- x[54],w[54][0],
- x[55],w[55][0],
- x[56],w[56][0],
- x[57],w[57][0],
- x[58],w[58][0],
- x[59],w[59][0],
- x[60],w[60][0],
- x[61],w[61][0],
- x[62],w[62][0],
- x[63],w[63][0]);
- out[1]=compute_engine64(x[0],w[0][1],
- x[1],w[1][1],
- x[2],w[2][1],
- x[3],w[3][1],
- x[4],w[4][1],
- x[5],w[5][1],
- x[6],w[6][1],
- x[7],w[7][1],
- x[8],w[8][1],
- x[9],w[9][1],
- x[10],w[10][1],
- x[11],w[11][1],
- x[12],w[12][1],
- x[13],w[13][1],
- x[14],w[14][1],
- x[15],w[15][1],
- x[16],w[16][1],
- x[17],w[17][1],
- x[18],w[18][1],
- x[19],w[19][1],
- x[20],w[20][1],
- x[21],w[21][1],
- x[22],w[22][1],
- x[23],w[23][1],
- x[24],w[24][1],
- x[25],w[25][1],
- x[26],w[26][1],
- x[27],w[27][1],
- x[28],w[28][1],
- x[29],w[29][1],
- x[30],w[30][1],
- x[31],w[31][1],
- x[32],w[32][1],
- x[33],w[33][1],
- x[34],w[34][1],
- x[35],w[35][1],
- x[36],w[36][1],
- x[37],w[37][1],
- x[38],w[38][1],
- x[39],w[39][1],
- x[40],w[40][1],
- x[41],w[41][1],
- x[42],w[42][1],
- x[43],w[43][1],
- x[44],w[44][1],
- x[45],w[45][1],
- x[46],w[46][1],
- x[47],w[47][1],
- x[48],w[48][1],
- x[49],w[49][1],
- x[50],w[50][1],
- x[51],w[51][1],
- x[52],w[52][1],
- x[53],w[53][1],
- x[54],w[54][1],
- x[55],w[55][1],
- x[56],w[56][1],
- x[57],w[57][1],
- x[58],w[58][1],
- x[59],w[59][1],
- x[60],w[60][1],
- x[61],w[61][1],
- x[62],w[62][1],
- x[63],w[63][1]);
-
-
-}
-
-void mvm512_512(ap_int<512> *data, float x[512],float mout[512],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
-#pragma HLS INTERFACE m_axi port=data
-#pragma HLS ARRAY_PARTITION variable=x cyclic dim=1 factor = 8
-#pragma HLS ARRAY_PARTITION variable=mout cyclic dim=1 factor = 8
- int i,j,k,l,kk;
- int index=0;
- for(k=0;k<512;k+=K_SIZE){
- /*loading*/
- for(i=0; i< K_SIZE;i++){
- #pragma HLS UNROLL factor=8
- kvector[i]=x[i+k];
- }
- for(l=0;l<512;l+=K_NUM){
-#pragma HLS pipeline ii=8
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][0]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][1]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
- kernel(kmatrix, kvector,kout);
- mout[l]+=kout[0];
- mout[l+1]+=kout[1];
-
- }
- }
-
-}
-
-void mvm512_1024(ap_int<512> *data, float x[512],float mout[1024],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
-#pragma HLS INTERFACE m_axi port=data
-#pragma HLS ARRAY_PARTITION variable=x cyclic dim=1 factor = 8
-#pragma HLS ARRAY_PARTITION variable=mout cyclic dim=1 factor = 8
- int i,j,k,l,kk;
- int index=0;
- for(k=0;k<512;k+=K_SIZE){
- /*loading*/
- for(i=0; i< K_SIZE;i++){
- #pragma HLS UNROLL factor=8
- kvector[i]=x[i+k];
- }
- for(l=0;l<1024;l+=K_NUM){
-#pragma HLS pipeline ii=8
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][0]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][1]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
- kernel(kmatrix, kvector,kout);
- mout[l]+=kout[0];
- mout[l+1]+=kout[1];
-
- }
- }
-
-}
-
-
-
-void mvm512_1024_2out(ap_int<512> *data, float x[512],float out1[1024],float out2[1024],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
-#pragma HLS INTERFACE m_axi port=data
-#pragma HLS ARRAY_PARTITION variable=x cyclic dim=1 factor = 8
-#pragma HLS ARRAY_PARTITION variable=out1 cyclic dim=1 factor = 8
-#pragma HLS ARRAY_PARTITION variable=out2 cyclic dim=1 factor = 8
- int i,j,k,l,kk;
- int index=0;
- for(k=0;k<512;k+=K_SIZE){
- /*loading*/
- for(i=0; i< K_SIZE;i++){
- #pragma HLS UNROLL factor=8
- kvector[i]=x[i+k];
- }
- for(l=0;l<1024;l+=K_NUM){
-#pragma HLS pipeline ii=8
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][0]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][1]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
- kernel(kmatrix, kvector,kout);
- out1[l]+=kout[0];
- out1[l+1]+=kout[1];
-
- }
- }
- for(k=0;k<512;k+=K_SIZE){
- /*loading*/
- for(i=0; i< K_SIZE;i++){
- #pragma HLS UNROLL factor=8
- kvector[i]=x[i+k];
- }
- for(l=0;l<1024;l+=K_NUM){
-#pragma HLS pipeline ii=8
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][0]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][1]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
- kernel(kmatrix, kvector,kout);
- out2[l]+=kout[0];
- out2[l+1]+=kout[1];
-
- }
- }
-
-
-}
-
-void mvm512_30000(ap_int<512> *data, float x[512],float mout[30000],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
-#pragma HLS INTERFACE m_axi port=data
-#pragma HLS ARRAY_PARTITION variable=x cyclic dim=1 factor = 8
-#pragma HLS ARRAY_PARTITION variable=mout cyclic dim=1 factor = 8
- int i,j,k,l,kk;
- int index=0;
- for(k=0;k<512;k+=K_SIZE){
- /*loading*/
- for(i=0; i< K_SIZE;i++){
- #pragma HLS UNROLL factor=8
- kvector[i]=x[i+k];
- }
- for(l=0;l<30000;l+=K_NUM){
-#pragma HLS pipeline ii=8
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][0]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][1]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
- kernel(kmatrix, kvector,kout);
- mout[l]+=kout[0];
- mout[l+1]+=kout[1];
-
- }
- }
-/*
- for(k=0;k<512;k+=K_SIZE){
-
- for(i=0; i< K_SIZE;i++){
- #pragma HLS UNROLL factor=8
- kvector[i]=x[i+k];
- }
-
- for(l=29952; l< 30000;l+=K_NUM){
-#pragma HLS pipeline
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][0]=trans_i(data[index].range(kk*32+31,kk*32);
- }
- index++;
- }
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][1]=trans_i(data[index].range(kk*32+31,kk*32);
- }
- index++;
- }
-
- kernel(kmatrix, kvector,kout);
- mout[l]+=kout[0];
- mout[l+1]+=kout[1];
-
- }
-
-
- }
- */
-}
-
-void mvm1024_512(ap_int<512> *data, float x[1024],float mout[512],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
-#pragma HLS INTERFACE m_axi port=data
-#pragma HLS ARRAY_PARTITION variable=x cyclic dim=1 factor = 8
-#pragma HLS ARRAY_PARTITION variable=mout cyclic dim=1 factor = 8
- int i,j,k,l,kk;
- int index=0;
- for(k=0;k<1024;k+=K_SIZE){
- /*loading*/
- for(i=0; i< K_SIZE;i++){
- #pragma HLS UNROLL factor=8
- kvector[i]=x[i+k];
- }
- for(l=0;l<512;l+=K_NUM){
-#pragma HLS pipeline ii=8
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][0]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][1]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
- kernel(kmatrix, kvector,kout);
- mout[l]+=kout[0];
- mout[l+1]+=kout[1];
-
- }
-
- }
-
-}
-
-void mvm1024_1024(ap_int<512> *data, float x[1024],float mout[1024],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
-#pragma HLS INTERFACE m_axi port=data
-#pragma HLS ARRAY_PARTITION variable=x cyclic dim=1 factor = 8
-#pragma HLS ARRAY_PARTITION variable=mout cyclic dim=1 factor = 8
- int i,j,k,l,kk;
- int index=0;
- for(k=0;k<1024;k+=K_SIZE){
- /*loading*/
- for(i=0; i< K_SIZE;i++){
- #pragma HLS UNROLL factor=8
- kvector[i]=x[i+k];
- }
- for(l=0;l<1024;l+=K_NUM){
-#pragma HLS pipeline ii=8
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][0]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][1]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
- kernel(kmatrix, kvector,kout);
- mout[l]+=kout[0];
- mout[l+1]+=kout[1];
-
- }
- }
-
-}
-
-void mvm1024_1024_2out(ap_int<512> *data,float x[1024],float out1[1024], float out2[1024],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
-#pragma HLS INTERFACE m_axi port=data
-#pragma HLS ARRAY_PARTITION variable=x cyclic dim=1 factor = 8
-#pragma HLS ARRAY_PARTITION variable=out1 cyclic dim=1 factor = 8
-#pragma HLS ARRAY_PARTITION variable=out2 cyclic dim=1 factor = 8
- int i,j,k,l,kk;
- int index=0;
- for(k=0;k<1024;k+=K_SIZE){
- /*loading*/
- for(i=0; i< K_SIZE;i++){
- #pragma HLS UNROLL factor=8
- kvector[i]=x[i+k];
- }
- for(l=0;l<1024;l+=K_NUM){
-#pragma HLS pipeline ii=8
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][0]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][1]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
- kernel(kmatrix, kvector,kout);
- out1[l]+=kout[0];
- out1[l+1]+=kout[1];
-
- }
- //printf("kvec %f \n",kvector[1]);
- //printf("kmatrix %f \n",kmatrix[10][1]);
- //printf("kout %f \n",kout[1]);
- }
-
- for(k=0;k<1024;k+=K_SIZE){
- /*loading*/
- for(i=0; i< K_SIZE;i++){
- #pragma HLS UNROLL factor=8
- kvector[i]=x[i+k];
- }
- for(l=0;l<1024;l+=K_NUM){
-#pragma HLS pipeline ii=8
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][0]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][1]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
- kernel(kmatrix, kvector,kout);
- out2[l]+=kout[0];
- out2[l+1]+=kout[1];
-
- }
-
- }
-
-}
-
-void mvm1024_1024_2in(ap_int<512> *data, float x1[1024],float x2[1024], float mout[1024],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
-#pragma HLS INTERFACE m_axi port=data
-#pragma HLS ARRAY_PARTITION variable=x1 cyclic dim=1 factor = 8
-#pragma HLS ARRAY_PARTITION variable=x2 cyclic dim=1 factor = 8
-#pragma HLS ARRAY_PARTITION variable=mout cyclic dim=1 factor = 8
- int i,j,k,l,kk;
- int index=0;
- for(k=0;k<1024;k+=K_SIZE){
- /*loading*/
- for(i=0; i< K_SIZE;i++){
- #pragma HLS UNROLL factor=8
- kvector[i]=x1[i+k];
- }
- for(l=0;l<1024;l+=K_NUM){
-#pragma HLS pipeline ii=8
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][0]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][1]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
- kernel(kmatrix, kvector,kout);
- mout[l]+=kout[0];
- mout[l+1]+=kout[1];
-
- }
-
- }
-
- for(k=0;k<1024;k+=K_SIZE){
- /*loading*/
- for(i=0; i< K_SIZE;i++){
- #pragma HLS UNROLL factor=8
- kvector[i]=x2[i+k];
- }
- for(l=0;l<1024;l+=K_NUM){
-#pragma HLS pipeline ii=8
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][0]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][1]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
- kernel(kmatrix, kvector,kout);
- mout[l]+=kout[0];
- mout[l+1]+=kout[1];
-
- }
-
- }
-}
-
-void mvm1024_2048(ap_int<512> *data, float x[1024],float mout[2048],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
-#pragma HLS INTERFACE m_axi port=data
-#pragma HLS ARRAY_PARTITION variable=x cyclic dim=1 factor = 8
-#pragma HLS ARRAY_PARTITION variable=mout cyclic dim=1 factor = 8
- int i,j,k,l,kk;
- int index=0;
- for(k=0;k<1024;k+=K_SIZE){
- /*loading*/
- for(i=0; i< K_SIZE;i++){
- #pragma HLS UNROLL factor=8
- kvector[i]=x[i+k];
- }
- for(l=0;l<2048;l+=K_NUM){
-#pragma HLS pipeline ii=8
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][0]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][1]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
- kernel(kmatrix, kvector,kout);
- mout[l]+=kout[0];
- mout[l+1]+=kout[1];
-
- }
- }
-
-}
-
-void mvm1024_2048_2in(ap_int<512> *data, float x1[1024],float x2[1024], float mout[2048],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
-#pragma HLS INTERFACE m_axi port=data
-#pragma HLS ARRAY_PARTITION variable=x1 cyclic dim=1 factor = 8
-#pragma HLS ARRAY_PARTITION variable=x2 cyclic dim=1 factor = 8
-#pragma HLS ARRAY_PARTITION variable=mout cyclic dim=1 factor = 8
- int i,j,k,l,kk;
- int index=0;
- for(k=0;k<1024;k+=K_SIZE){
- /*loading*/
- for(i=0; i< K_SIZE;i++){
- #pragma HLS UNROLL factor=8
- kvector[i]=x1[i+k];
- }
- for(l=0;l<2048;l+=K_NUM){
-#pragma HLS pipeline ii=8
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][0]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][1]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
- kernel(kmatrix, kvector,kout);
- mout[l]+=kout[0];
- mout[l+1]+=kout[1];
-
- }
- }
-
- for(k=0;k<1024;k+=K_SIZE){
- /*loading*/
- for(i=0; i< K_SIZE;i++){
- #pragma HLS UNROLL factor=8
- kvector[i]=x2[i+k];
- }
- for(l=0;l<2048;l+=K_NUM){
-#pragma HLS pipeline ii=8
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][0]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][1]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
- kernel(kmatrix, kvector,kout);
- mout[l]+=kout[0];
- mout[l+1]+=kout[1];
-
- }
- }
-}
-void mvm2048_512(ap_int<512> *data, float x[2048],float mout[512],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
-#pragma HLS INTERFACE m_axi port=data
-#pragma HLS ARRAY_PARTITION variable=x cyclic dim=1 factor = 8
-#pragma HLS ARRAY_PARTITION variable=mout cyclic dim=1 factor = 8
- int i,j,k,l,kk;
- int index=0;
- for(k=0;k<2048;k+=K_SIZE){
- /*loading*/
- for(i=0; i< K_SIZE;i++){
- #pragma HLS UNROLL factor=8
- kvector[i]=x[i+k];
- }
- for(l=0;l<512;l+=K_NUM){
-#pragma HLS pipeline ii=8
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][0]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][1]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
- kernel(kmatrix, kvector,kout);
- mout[l]+=kout[0];
- mout[l+1]+=kout[1];
-
- }
- }
-
-}
-
-void mvm2048_1024(ap_int<512> *data, float x[2048],float mout[1024],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
-#pragma HLS INTERFACE m_axi port=data
-#pragma HLS ARRAY_PARTITION variable=x cyclic dim=1 factor = 8
-#pragma HLS ARRAY_PARTITION variable=mout cyclic dim=1 factor = 8
- int i,j,k,l,kk;
- int index=0;
- for(k=0;k<2048;k+=K_SIZE){
- /*loading*/
- for(i=0; i< K_SIZE;i++){
- #pragma HLS UNROLL factor=8
- kvector[i]=x[i+k];
- }
- for(l=0;l<1024;l+=K_NUM){
-#pragma HLS pipeline ii=8
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][0]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][1]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
- kernel(kmatrix, kvector,kout);
- mout[l]+=kout[0];
- mout[l+1]+=kout[1];
-
- }
- }
-
-}
-
-void mvm2048_1024_2out(ap_int<512> *data, float x[2048],float out1[1024],float out2[1024],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
-#pragma HLS INTERFACE m_axi port=data
-#pragma HLS ARRAY_PARTITION variable=x cyclic dim=1 factor = 8
-#pragma HLS ARRAY_PARTITION variable=out1 cyclic dim=1 factor = 8
-#pragma HLS ARRAY_PARTITION variable=out2 cyclic dim=1 factor = 8
- int i,j,k,l,kk;
- int index=0;
- for(k=0;k<2048;k+=K_SIZE){
- /*loading*/
- for(i=0; i< K_SIZE;i++){
- #pragma HLS UNROLL factor=8
- kvector[i]=x[i+k];
- }
- for(l=0;l<1024;l+=K_NUM){
-#pragma HLS pipeline ii=8
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][0]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][1]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
- kernel(kmatrix, kvector,kout);
- out1[l]+=kout[0];
- out1[l+1]+=kout[1];
-
- }
- }
-
- for(k=0;k<2048;k+=K_SIZE){
- /*loading*/
- for(i=0; i< K_SIZE;i++){
- #pragma HLS UNROLL factor=8
- kvector[i]=x[i+k];
- }
- for(l=0;l<1024;l+=K_NUM){
-#pragma HLS pipeline ii=8
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][0]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][1]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
- kernel(kmatrix, kvector,kout);
- out2[l]+=kout[0];
- out2[l+1]+=kout[1];
-
- }
- }
-}
-
-void mvm2048_2048(ap_int<512> *data, float x[2048],float mout[2048],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
-#pragma HLS INTERFACE m_axi port=data
-#pragma HLS ARRAY_PARTITION variable=x cyclic dim=1 factor = 8
-#pragma HLS ARRAY_PARTITION variable=mout cyclic dim=1 factor = 8
- int i,j,k,l,kk;
- int index=0;
- for(k=0;k<2048;k+=K_SIZE){
- /*loading*/
- for(i=0; i< K_SIZE;i++){
- #pragma HLS UNROLL factor=8
- kvector[i]=x[i+k];
- }
- for(l=0;l<2048;l+=K_NUM){
-#pragma HLS pipeline ii=8
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][0]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
- for(j=0;j< K_SIZE; j+=16){
- for (kk=0; kk<16; kk++){
- kmatrix[j+kk][1]=trans_i(data[index].range(kk*32+31,kk*32));
- }
- index++;
- }
-
- kernel(kmatrix, kvector,kout);
- mout[l]+=kout[0];
- mout[l+1]+=kout[1];
-
- }
- }
-
-}
diff --git a/nmtdecoder/mvm.h b/nmtdecoder/mvm.h
deleted file mode 100644
index fc97d4dae37e6b9e4d725d689f3a7311806ae4c0..0000000000000000000000000000000000000000
--- a/nmtdecoder/mvm.h
+++ /dev/null
@@ -1,43 +0,0 @@
-#ifndef MVM_H
-#define MVM_H
-#include "ce.h"
-#define K_SIZE 64
-#define K_NUM 2
-#include "ap_fixed.h"
-typedef union {
- unsigned int i;
- float f;
- } u;
-
-float trans_i(ap_int<32> a);
-void kernel(float w[K_SIZE][K_NUM], float x[K_SIZE],float out[K_NUM]);
-
-void mvm512_512(ap_int<512> *data, float x[512],float mout[512],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-
-void mvm512_1024(ap_int<512> *data, float x[512],float mout[1024],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-
-void mvm512_1024_2out(ap_int<512> *data, float x[512],float out1[1024],float out2[1024],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-
-void mvm512_30000(ap_int<512> *data, float x[512],float mout[30000],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-
-void mvm1024_512(ap_int<512> *data, float x[1024],float mout[512],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-
-void mvm1024_1024(ap_int<512> *data, float x[1024],float mout[1024],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-
-void mvm1024_1024_2out(ap_int<512> *data, float x[1024],float out1[1024], float out2[1024],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-
-void mvm1024_1024_2in(ap_int<512> *data, float x1[1024],float x2[1024], float mout[1024],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-
-void mvm1024_2048(ap_int<512> *data, float x[1024],float mout[2048],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-
-void mvm1024_2048_2in(ap_int<512> *data, float x1[1024],float x2[1024], float mout[2048],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-
-void mvm2048_512(ap_int<512> *data, float x[2048],float mout[512],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-
-void mvm2048_1024(ap_int<512> *data, float x[2048],float mout[1024],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-
-void mvm2048_1024_2out(ap_int<512> *data, float x[2048],float out1[1024],float out2[1024],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-
-void mvm2048_2048(ap_int<512> *data, float x[2048],float mout[2048],float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-
-#endif
diff --git a/nmtdecoder/nmtdecoder.cc b/nmtdecoder/nmtdecoder.cc
deleted file mode 100644
index 3ff608ec9c819937ea8139eb0f62865004706416..0000000000000000000000000000000000000000
--- a/nmtdecoder/nmtdecoder.cc
+++ /dev/null
@@ -1,76 +0,0 @@
-#include "nmtdecoder.h"
-//#include <stdio.h>
-
-void ff_group(
- ap_int<512> *data,
- float input1[dim], float output1[dim_word],
- float input2[dim_word], float output2[dim_word],
- float input3[ctxdim], float output3[dim_word],
- float input[dim_word], float output[n_words],
- float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]){
-#pragma HLS inline off
-#pragma HLS inline region recursive
-//#pragma HLS ALLOCATION instances=compute_engine64 limit=1 function
- #pragma HLS INTERFACE m_axi port=data
- int i ;
- logit_lstm(data,input1, output1,kvector,kmatrix,kout);
-
- //printf("\nlstm %f\n",output1[10]);
- logit_prev(data+32800,input2, output2,kvector,kmatrix,kout);
- //printf("prev %f\n",output2[10]);
-
- logit_ctx(data+49216,input3, output3,kvector,kmatrix,kout);
- //printf("ctx %f\n",output3[10]);
-
- for(i =0 ; i< dim_word; i++){
-//#pragma HLS unroll factor =16
- input[i] = tanh_hls((float)(output1[i]+ output2[i] +output3[i]));
- }
-
- logit(data+114784, input, output,kvector,kmatrix,kout);
- //printf("logit %f\n",output[10]);
- };
-
-
-void nmtdecoder(
- ap_int<512> *data,
- int stepsize,
- float encoder_output[MAXSTEP][dim],
- float encoder_r_output[MAXSTEP][dim],
- float memory[dim],
- float de_input[dim_word],
- float decoder_output[dim],
- float logit_output[n_words],
- float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]
- ){
-
-#pragma HLS inline off
-#pragma HLS inline region recursive
-//#pragma HLS ALLOCATION instances=compute_engine64 limit=1 function
- #pragma HLS INTERFACE m_axi port=data
- float attention_output[ctxdim];
- float lstm_output[dim_word];
- float prev_output[dim_word];
- float ctx_output[dim_word];
- float decoder_output1[dim];
- float logit_input[dim_word];
- int i=0;
- gru_layer_base(data,de_input,decoder_output1,memory,kvector,kmatrix,kout);
- //printf("\ndecoder output 1 %f %f %f %f %f %f\n",decoder_output1[0],decoder_output1[50],decoder_output1[100],decoder_output1[500],decoder_output1[800],decoder_output1[1000]);
- //for(i =0;i<1024;i++)
- //printf("decoder output 1 %f ",decoder_output1[i]);
- attention(data+295104,decoder_output1,encoder_output,encoder_r_output,stepsize,attention_output,kvector,kmatrix,kout);
- //for(i =0;i<1024;i++)
- //printf("Attention output %f ",attention_output[i]);
- //printf("Attention output %f %f %f %f %f %f\n",attention_output[0],attention_output[100],attention_output[200],attention_output[1000],attention_output[1500],attention_output[2000]);
- gru_layer_nl(data+688577,attention_output,decoder_output,decoder_output1,kvector,kmatrix,kout);
- //printf("decoder output 2 %f %f %f %f %f %f\n",decoder_output[0],decoder_output[100],decoder_output[102],decoder_output[200],decoder_output[601],decoder_output[1023]);
- ff_group(data+1278593,decoder_output, lstm_output,de_input, prev_output,
- attention_output, ctx_output,
- logit_input,logit_output,kvector,kmatrix,kout);
- //printf("logit output %f %f %f %f %f %f\n",logit_output[0],logit_output[511],logit_output[10000],logit_output[15000],logit_output[20000],logit_output[29000]);
-
-}
-
-
-
diff --git a/nmtdecoder/nmtdecoder.h b/nmtdecoder/nmtdecoder.h
deleted file mode 100644
index 24694d42203035437d491719c2f9c53062d90c93..0000000000000000000000000000000000000000
--- a/nmtdecoder/nmtdecoder.h
+++ /dev/null
@@ -1,24 +0,0 @@
-
-#ifndef NMTDECODER_H
-#define NMTDECODER_H
-#include "layers.h"
-void ff_group(
- ap_int<512> *data,
- float input1[dim], float output1[dim_word],
- float input2[dim_word], float output2[dim_word],
- float input3[ctxdim], float output3[dim_word],
- float input[dim_word], float output[n_words],
- float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]);
-void nmtdecoder(
- ap_int<512> *data,
- int stepsize,
- float encoder_output[MAXSTEP][dim],
- float encoder_r_output[MAXSTEP][dim],
- float memory[dim],
- float de_input[dim_word],
- float decoder_output[dim],
- float logit_output[n_words],
- float kvector[K_SIZE],float kmatrix[K_SIZE][K_NUM],float kout[K_NUM]
- );
-
-#endif
diff --git a/DRED_codegen/q_learning.py b/q_learning.py
similarity index 100%
rename from DRED_codegen/q_learning.py
rename to q_learning.py
diff --git a/DRED_codegen/q_learning_v2.py b/q_learning_v2.py
similarity index 100%
rename from DRED_codegen/q_learning_v2.py
rename to q_learning_v2.py
diff --git a/DRED_codegen/saved_networks_q_learning/-dqn-8340.data-00000-of-00001 b/saved_networks_q_learning/-dqn-8340.data-00000-of-00001
similarity index 100%
rename from DRED_codegen/saved_networks_q_learning/-dqn-8340.data-00000-of-00001
rename to saved_networks_q_learning/-dqn-8340.data-00000-of-00001
diff --git a/DRED_codegen/saved_networks_q_learning/-dqn-8340.index b/saved_networks_q_learning/-dqn-8340.index
similarity index 100%
rename from DRED_codegen/saved_networks_q_learning/-dqn-8340.index
rename to saved_networks_q_learning/-dqn-8340.index
diff --git a/DRED_codegen/saved_networks_q_learning/-dqn-8340.meta b/saved_networks_q_learning/-dqn-8340.meta
similarity index 100%
rename from DRED_codegen/saved_networks_q_learning/-dqn-8340.meta
rename to saved_networks_q_learning/-dqn-8340.meta
diff --git a/DRED_codegen/saved_networks_q_learning/-dqn-8350.data-00000-of-00001 b/saved_networks_q_learning/-dqn-8350.data-00000-of-00001
similarity index 100%
rename from DRED_codegen/saved_networks_q_learning/-dqn-8350.data-00000-of-00001
rename to saved_networks_q_learning/-dqn-8350.data-00000-of-00001
diff --git a/DRED_codegen/saved_networks_q_learning/-dqn-8350.index b/saved_networks_q_learning/-dqn-8350.index
similarity index 100%
rename from DRED_codegen/saved_networks_q_learning/-dqn-8350.index
rename to saved_networks_q_learning/-dqn-8350.index
diff --git a/DRED_codegen/saved_networks_q_learning/-dqn-8350.meta b/saved_networks_q_learning/-dqn-8350.meta
similarity index 100%
rename from DRED_codegen/saved_networks_q_learning/-dqn-8350.meta
rename to saved_networks_q_learning/-dqn-8350.meta
diff --git a/DRED_codegen/saved_networks_q_learning/-dqn-8360.data-00000-of-00001 b/saved_networks_q_learning/-dqn-8360.data-00000-of-00001
similarity index 100%
rename from DRED_codegen/saved_networks_q_learning/-dqn-8360.data-00000-of-00001
rename to saved_networks_q_learning/-dqn-8360.data-00000-of-00001
diff --git a/DRED_codegen/saved_networks_q_learning/-dqn-8360.index b/saved_networks_q_learning/-dqn-8360.index
similarity index 100%
rename from DRED_codegen/saved_networks_q_learning/-dqn-8360.index
rename to saved_networks_q_learning/-dqn-8360.index
diff --git a/DRED_codegen/saved_networks_q_learning/-dqn-8360.meta b/saved_networks_q_learning/-dqn-8360.meta
similarity index 100%
rename from DRED_codegen/saved_networks_q_learning/-dqn-8360.meta
rename to saved_networks_q_learning/-dqn-8360.meta
diff --git a/DRED_codegen/saved_networks_q_learning/-dqn-8370.data-00000-of-00001 b/saved_networks_q_learning/-dqn-8370.data-00000-of-00001
similarity index 100%
rename from DRED_codegen/saved_networks_q_learning/-dqn-8370.data-00000-of-00001
rename to saved_networks_q_learning/-dqn-8370.data-00000-of-00001
diff --git a/DRED_codegen/saved_networks_q_learning/-dqn-8370.index b/saved_networks_q_learning/-dqn-8370.index
similarity index 100%
rename from DRED_codegen/saved_networks_q_learning/-dqn-8370.index
rename to saved_networks_q_learning/-dqn-8370.index
diff --git a/DRED_codegen/saved_networks_q_learning/-dqn-8370.meta b/saved_networks_q_learning/-dqn-8370.meta
similarity index 100%
rename from DRED_codegen/saved_networks_q_learning/-dqn-8370.meta
rename to saved_networks_q_learning/-dqn-8370.meta
diff --git a/DRED_codegen/saved_networks_q_learning/-dqn-8380.data-00000-of-00001 b/saved_networks_q_learning/-dqn-8380.data-00000-of-00001
similarity index 100%
rename from DRED_codegen/saved_networks_q_learning/-dqn-8380.data-00000-of-00001
rename to saved_networks_q_learning/-dqn-8380.data-00000-of-00001
diff --git a/DRED_codegen/saved_networks_q_learning/-dqn-8380.index b/saved_networks_q_learning/-dqn-8380.index
similarity index 100%
rename from DRED_codegen/saved_networks_q_learning/-dqn-8380.index
rename to saved_networks_q_learning/-dqn-8380.index
diff --git a/DRED_codegen/saved_networks_q_learning/-dqn-8380.meta b/saved_networks_q_learning/-dqn-8380.meta
similarity index 100%
rename from DRED_codegen/saved_networks_q_learning/-dqn-8380.meta
rename to saved_networks_q_learning/-dqn-8380.meta
diff --git a/DRED_codegen/saved_networks_q_learning/checkpoint b/saved_networks_q_learning/checkpoint
similarity index 100%
rename from DRED_codegen/saved_networks_q_learning/checkpoint
rename to saved_networks_q_learning/checkpoint
diff --git a/DRED_codegen/script.tcl b/script.tcl
similarity index 100%
rename from DRED_codegen/script.tcl
rename to script.tcl
diff --git a/DRED_codegen/synth.log b/synth.log
similarity index 100%
rename from DRED_codegen/synth.log
rename to synth.log
diff --git a/DRED_codegen/tempate.py b/tempate.py
similarity index 100%
rename from DRED_codegen/tempate.py
rename to tempate.py
diff --git a/DRED_codegen/templates/nmt.jinja b/templates/nmt.jinja
similarity index 100%
rename from DRED_codegen/templates/nmt.jinja
rename to templates/nmt.jinja
diff --git a/DRED_codegen/templates/tcl_script.jinja b/templates/tcl_script.jinja
similarity index 100%
rename from DRED_codegen/templates/tcl_script.jinja
rename to templates/tcl_script.jinja
diff --git a/DRED_codegen/vivado_hls.log b/vivado_hls.log
similarity index 100%
rename from DRED_codegen/vivado_hls.log
rename to vivado_hls.log