Add bug-fixed version of quantized convolution IP with kn2row and tiling

dlk/backends/.clang-format

@@ -0,0 +1,42 @@
+BasedOnStyle: 'Google'
+AccessModifierOffset: -2
+AlignEscapedNewlinesLeft: true
+AlignConsecutiveAssignments: false
+AlignEscapedNewlinesLeft: true
+AlignTrailingComments: true
+AllowShortBlocksOnASingleLine: true
+AllowShortCaseLabelsOnASingleLine: false
+AllowShortFunctionsOnASingleLine: true
+AllowShortIfStatementsOnASingleLine: false
+AllowShortLoopsOnASingleLine: true
+AlwaysBreakAfterDefinitionReturnType: false
+AlwaysBreakBeforeMultilineStrings: true
+AlwaysBreakTemplateDeclarations: true
+BinPackArguments: true
+BinPackParameters: true
+BreakBeforeBinaryOperators: 'None'
+BreakBeforeBraces: 'Mozilla'
+BreakConstructorInitializersBeforeComma: true
+ColumnLimit: 120
+ConstructorInitializerAllOnOneLineOrOnePerLine: true
+ConstructorInitializerIndentWidth: 2
+ContinuationIndentWidth: 2
+Cpp11BracedListStyle: true
+DerivePointerBinding: true
+IndentCaseLabels: true
+IndentWidth: 2
+IndentWrappedFunctionNames: false
+KeepEmptyLinesAtTheStartOfBlocks: true
+MaxEmptyLinesToKeep: 1
+NamespaceIndentation: 'None'
+PointerAlignment: 'Left'
+SpaceBeforeAssignmentOperators: true
+SpaceBeforeParens: 'ControlStatements'
+SpaceInEmptyParentheses: false
+SpacesBeforeTrailingComments: 1
+SpacesInAngles: false
+SpacesInCStyleCastParentheses: false
+SpacesInParentheses: false
+Standard: 'Cpp03'
+TabWidth: 2
+UseTab: 'Never'

dlk/backends/CMakeLists.txt

@@ -1,11 +1,9 @@
 cmake_minimum_required(VERSION 3.1)
-project(conv_sample CXX)
+project(dlk_backends CXX)
 
 
 set(CMAKE_CXX_STANDARD_REQUIRED ON)
 set(CMAKE_CXX_EXTENSIONS OFF)
-set(CMAKE_BUILD_TYPE Debug)
-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ")
 
 get_filename_component(ROOT_DIR ${CMAKE_SOURCE_DIR} ABSOLUTE)
 get_filename_component(INTEL_HLS_DIR "${SRC_DIR}/intel_hls" ABSOLUTE)
@@ -29,25 +27,33 @@ elseif(UNIX AND NOT APPLE)
     get_filename_component(INTEL_HLS_LIB_DIR "$ENV{HLS_INSTALL_DIR}/host/linux64/lib/" ABSOLUTE)
 endif()
 
-
 include_directories(${INCLUDE_DIR})
-# link_directories(${HLS_LIB_DIR})
 
 file(GLOB_RECURSE TB_SRC "${TB_DIR}/*.cpp")
 file(GLOB_RECURSE CPP_SRC "${CPP_DIR}/*.cpp")
 file(GLOB_RECURSE INTEL_HLS_SRC "${INTEL_HLS_DIR}/*.cpp")
 set(SRC ${TB_SRC} ${CPP_SRC})
 
+enable_testing() 
 add_executable(x86 ${SRC})
 target_compile_options(x86 PRIVATE -std=c++0x
+                                   -O0
                                    -Wall
                                    -Wextra
                                    -Werror
                                    -Wno-unknown-pragmas
                                    -Wno-unused-function
                                    -Wno-unused-parameter
                                    -Wno-unused-variable
-                                   -Wno-sign-compare)
+                                   -Wno-sign-compare
+                                   -fno-omit-frame-pointer
+                                   -fsanitize=address)
+target_link_libraries(x86
+  -fsanitize=address
+)
+add_test(NAME x86_test 
+         COMMAND x86 random
+         WORKING_DIRECTORY ${CMAKE_BINARY_DIR})
 
 
 if(UNIX AND NOT APPLE)

dlk/backends/Makefile

@@ -1,12 +1,8 @@
 DEVICE := 5CSEBA6U23I7
-CLOCK  := 50MHz
+CLOCK  := 75MHz
 
 CXX :=
-# COMPONENTS := --component intel_hls_a8w1_qconv_with_kn2row_impl,intel_hls_qconv_with_kn2row_impl
-# COMPONENTS := --component intel_hls_qconv_with_kn2row_impl
-# COMPONENTS := --component intel_hls_apply_thresholds_impl
-COMPONENTS := --component intel_hls_qconv_with_kn2row_impl,intel_hls_apply_thresholds_impl
-# COMPONENTS := --component intel_hls_a8w1_qconv_with_kn2row_impl
+COMPONENTS := --component intel_hls_qconv_kn2row_tiling_impl
 
 SRC_DIR := ./src
 TB_DIR := $(SRC_DIR)/tb
@@ -52,13 +48,13 @@ clean:
 
 
 x86:	CXX = g++
-x86:	FLAGS += $(INCLUDE) -O3 -std=c++0x -g
+x86:	FLAGS += $(INCLUDE) -O0 -std=c++0x -g
 
 arm:	CXX = arm-linux-gnueabihf-g++
 arm:    FLAGS += $(INCLUDE) -static -s -std=c++0x -O3
 
 fpga:	CXX = arm-linux-gnueabihf-g++
-fpga:	FLAGS += $(INCLUDE) -static -s -std=c++0x -O3 -D_DE10_NANO_
+fpga:	FLAGS += $(INCLUDE) -static -std=c++0x -s -O3 -D_DE10_NANO_
 
 sim:   CXX = i++
 sim:   FLAGS += -march=x86-64 $(INTEL_HLS_INCLUDE) -lm $(COMPONENTS) -D_INTEL_HLS_

dlk/backends/include/common/global.h

@@ -14,20 +14,20 @@ limitations under the License.
 ==============================================================================*/
 
 #pragma once
-#include <iostream>
-#include <cmath>
 #include <assert.h>
+#include <cmath>
 #include <cstring>
+#include <iostream>
 #include "common/type_info.h"
 
 #include "params/conv.h"
 #include "params/gemm.h"
 
 #define IP_CSR_ADDR 0xFF200000
 #define A8W1_IP_CSR_ADDR 0xFF200100
-#define IN_DATA_ADDR  0x20000000
+#define IN_DATA_ADDR 0x20000000
 #define OUT0_DATA_ADDR 0x2C000000
 #define OUT1_DATA_ADDR 0x32000000
 #define OUT_DATA_ADDR OUT0_DATA_ADDR
-#define K_DATA_ADDR   0x38000000
-#define THRESHOLDS_ADDR 0x3F00000
+#define K_DATA_ADDR 0x38000000
+#define THRESHOLDS_ADDR 0x3F000000