Batch and SVM (#2)

* toggle interleaving using cmd line args, svm init

* modified tests based on interleaving args

* enabled fast emulation

* Working svm host code

* Buffer locations attribute in fft3d kernel

* working batch fft3d svm

* working 3D FFT with bitrev diagonal transpose

* loop coalesced fft kernels

* fft3d: single loop dbl buffered fetch

* fft3d: fetch_transpose working

* Buffer location in stores

- buffer location to enable DDR stores in SVM,

- print iteration wise performance measurments

- modify kernel names in host to match the kernel

* FFT2d: bram dbl buf and single loop

* correct svm buffer flags, init output buffer, fft1d iter

* clFinish stms in DDR buffer reads

* FFT3D: separated svm and ddr

* FFT3D: Working opt bram transpose, bool instead of int

* FFT3D: renamed bram transpose files, dependencies

* CI: fixed host executable names

* FFT1d: working svm

* CI: corrected build, artifact paths

* Test: fix path to 2d bram

* CI: Adding release debug to restrict prints

* Test: fixed path for 3d FFT DDR

* modified verification function

* Working DDR batch of 2

* working DDR batch for all

* Split to individual files

* Hw Counters and const inp

* Reordered host kernel calls, removed channel depth

* batched fft3d with safelen

* fixed host code for svm, svm batch

* working svm batch

* no verify param

* free forgotten

* porting example to cpp, using cxxopts, cmake files to folder

* removed argparse, refactored timing

* verify bitrev fft1d, using opt bram fft3d

* working ddr batch

* streamlined synth path

* identical fft3d ddr and svm kernels

* updated tests

* removed extern modules

* removed common cl files to common dir

* gitlab ci update

* updated readme

* updated changelog

* fixed args in fft1d, fft2d

* fixed ddr batch and renamed queues

* added additional variants to application selection

* expressive emulation and report generation

* cleaned up application files

* fixed emulation dependency in test

* draft userguide

* fixed gitlab ci

* added power scripts

* minor change

* updated changelog

* ci: test fft1d

* updated ci with 2d, 3d variants

* removed fft3d bram due to lack of cpu resources

* cmake build type selection

* removed caching cmake variables for sdk and bsp

* modified readme and updated userguide

Authored-by: Arjun Ramaswami <[email protected]>

Author: arjunramaswami

.gitignore

@@ -1,11 +1,15 @@
-build/
-debug/
+build*/
+debug*/
 test/
 bin/
 fpgabitstream/
 reports/
 vscode/
-scripts/
+scripts/slurm
+build*
+svm_build/
+debug*
+test_*
 
 tags
 *.DS_Store

.gitlab-ci.yml

@@ -4,36 +4,58 @@ variables:
 stages:
   - build
   - test
+  - test_fft1d
+  - test_fft2d_ddr
+  - test_fft2d_bram
+  - test_fft3d_ddr
 
 build-all:
   stage: build
   script:
     - rm -rf build
     - mkdir -p build && cd build
-    - cmake -DLOG_FFT_SIZE=6 ..
+    - srun -A pc2-mitarbeiter -p fpga --constraint=emul --pty bash
+    - cmake -DLOG_FFT_SIZE=6 -DCMAKE_BUILD_TYPE=Release ..
     - make
-    - make fft1d_emu
-    - make fft2d_ddr_emu
-    - make fft2d_bram_emu
-    - make fft3d_ddr_triv_emu
-    - make fft3d_bram_emu
-    - chmod +x bin/fft3d bin/fft2d bin/fft1d bin/test_fftfpga
+    - chmod +x bin/fft
   artifacts:
     paths:
-      - build/bin/emu_64_fft1d/fft1d.aocx
-      - build/bin/emu_64_fft2d_ddr/fft2d_bram.aocx
-      - build/bin/emu_64_fft2d_bram/fft2d_ddr.aocx
-      - build/bin/emu_64_fft3d_ddr_triv/fft3d_ddr_triv.aocx
-      - build/bin/emu_64_fft3d_bram/fft3d_bram.aocx
+      - build/bin/p520_hpc_sg280l/emulation/fft1d_64_nointer/fft1d.aocx
+      - build/bin/p520_hpc_sg280l/emulation/fft2d_bram_64_nointer/fft2d_bram.aocx
+      - build/bin/p520_hpc_sg280l/emulation/fft2d_ddr_64_nointer/fft2d_ddr.aocx
+      - build/bin/p520_hpc_sg280l/emulation/fft3d_ddr_64_nointer/fft3d_ddr.aocx
+      - build/bin/p520_hpc_sg280l/emulation/fft3d_bram_64_nointer/fft3d_bram.aocx
       - build/bin/test_fftfpga
-      - build/bin/fft3d
-      - build/bin/fft2d
-      - build/bin/fft1d
+      - build/bin/fft
 
 test-all:
   stage: test
   script:
     - cd build/bin/
-    - CL_CONTEXT_EMULATOR_DEVICE_INTELFPGA=1 ./test_fftfpga
+    - CL_CONFIG_CPU_EMULATE_DEVICES=1 ./test_fftfpga
   dependencies:
-    - build-all
+    - build-all
+
+test-fft1d:
+  stage: test_fft1d
+  script:
+    - cd build/bin/
+    - ./fft -n 64 -d 1 --emulate -p p520_hpc_sg280l/emulation/fft1d_64_nointer/fft1d.aocx
+
+test-fft2d_ddr:
+  stage: test_fft2d_ddr
+  script:
+    - cd build/bin/
+    - ./fft -n 64 -d 2 --emulate -p p520_hpc_sg280l/emulation/fft2d_ddr_64_nointer/fft2d_ddr.aocx
+
+test-fft2d_bram:
+  stage: test_fft2d_bram
+  script:
+    - cd build/bin/
+    - ./fft -n 64 -d 2 --emulate --use_bram -p p520_hpc_sg280l/emulation/fft2d_bram_64_nointer/fft2d_bram.aocx
+
+test-fft3d_ddr:
+  stage: test_fft3d_ddr
+  script:
+    - cd build/bin/
+    - ./fft -n 64 -d 3 --emulate -p p520_hpc_sg280l/emulation/fft3d_ddr_64_nointer/fft3d_ddr.aocx

.gitmodules

@@ -1,23 +1,15 @@
 # Changelog
+
 All notable changes to this project will be documented in this file.
 
 ## Unreleased
-- coarse grained SVM support for Intel FPGAs
-- host code as library
-- configurable platform name
-- choice of platform (currently chooses first platform)
-- choice of device (currently chooses first device)
-- loads binary to multiple devices (currently only the first device)
-- batch mode
-- xilinx fpgas
 
-## [1.0.0] - []
+- configurable CL platform and device
+
+## [1.0.0] - [18.10.2021]
 
-### Added
-- 3d FFT in 2 varients, in bram and ddr transpose
-- 2d FFT in 2 varients, in bram and ddr transpose
-- 1d FFT
-- CI/CD
-- gtests
-- CMake build
-- License
+- 1d, 2d, 3d FFT with variants on the location of the transposition
+- Batched 3D FFT to schedule multiple transformations by overlapping data transfers and FFT computations
+- Using OpenCL Shared Virtual Memory (SVM) for data transfers between FPGA and host
+- Unit tests
+- CMake build system

CHANGELOG

@@ -1,61 +1,40 @@
 # Author: Arjun Ramaswami
-cmake_minimum_required (VERSION 3.1)
-
-project(fft)
+cmake_minimum_required (VERSION 3.10)
+project(fft VERSION 2.0
+          DESCRIPTION "OpenCL based FFT library for Intel FPGAs"
+          LANGUAGES C CXX)
 
 set(CMAKE_C_STANDARD 11)
-
 set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${PROJECT_BINARY_DIR}/bin)
 
-# Setup submodules that are required
-find_package(Git QUIET)
-if(GIT_FOUND AND EXISTS "${PROJECT_SOURCE_DIR}/.git")
-# Update submodules as needed
-    option(GIT_SUBMODULE "Check submodules during build" ON)
-    if(GIT_SUBMODULE)
-        message(STATUS "Submodule update")
-        execute_process(COMMAND ${GIT_EXECUTABLE} submodule update --init --recursive
-                        WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}
-                        RESULT_VARIABLE GIT_SUBMOD_RESULT)
-        if(NOT GIT_SUBMOD_RESULT EQUAL "0")
-            message(FATAL_ERROR "git submodule update --init failed with ${GIT_SUBMOD_RESULT}, please checkout submodules")
-        endif()
-    endif()
-endif()
-
-# Include hlslib in CMake module path
-set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} ${CMAKE_SOURCE_DIR}/extern/hlslib/cmake)
+# build external dependencies
+message("-- Building external dependencies")
+include(${CMAKE_CURRENT_SOURCE_DIR}/cmake/extDep.cmake)
 
-# Find hlslib Intel OpenCL kernels
+## find Intel FPGA SDK for OpenCL
 find_package(IntelFPGAOpenCL REQUIRED)
-
-# Find FFTW
-set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} ${CMAKE_SOURCE_DIR}/extern/findFFTW)
-
-find_package(FFTW)
-if(FFTW_FOUND)
-  message("-- FFTW library found")
-else()
-  message(WARNING, "FFTW library not found")
+message("-- IntelFPGAOpenCL found")
+
+## find FFTW
+list(APPEND CMAKE_MODULE_PATH "${CMAKE_SOURCE_DIR}/cmake")
+find_package(FFTW REQUIRED)
+message("-- FFTW found")
+
+# Set a default build type if none was specified
+if(NOT CMAKE_BUILD_TYPE)
+  message(STATUS "Setting build type to 'Debug' as none was specified.")
+  set(CMAKE_BUILD_TYPE Debug CACHE STRING "Choose the type of build." FORCE)
+  # Set the possible values of build type for cmake-gui
+  set_property(CACHE CMAKE_BUILD_TYPE PROPERTY STRINGS "Debug" "Release"
+    "RelWithDebInfo")
 endif()
 
-# Link argparse as static library
-add_subdirectory(${CMAKE_SOURCE_DIR}/extern/argparse)
-
-# Add sub directories
+# sub directories
 add_subdirectory(api)
 add_subdirectory(kernels)
 add_subdirectory(examples)
 
-# GTest Setup
-if(CMAKE_PROJECT_NAME STREQUAL PROJECT_NAME)
-  option(PACKAGE_TESTS "Build the tests" ON)
-endif()
-
-if(PACKAGE_TESTS)
-    enable_testing()
-    include(GoogleTest)
-    add_subdirectory(tests)
-endif()
-
-# TODO: shift to external project instead of submodules as suggested by GTest
+# build tests
+message("-- Building tests")
+add_subdirectory(tests)
+enable_testing()