Add dav1d as potential external project

To add the external source, run
"git clone -b 1.5.0 https://code.videolan.org/videolan/dav1d.git"
to clone the dav1d sources, e.g. in the
llvm-test-suite/test-suite-externals directory (which may not exist
but can be created manually), or configure with
TEST_SUITE_DAV1D_ROOT pointing at a directory containing this source.

This builds two targets; the dav1d command line executable (which
isn't executed as a test) and the dav1d checkasm test executable
(which runs compiler generated functions and compares them with
handwritten assembly versions of them). The checkasm execuable
can also be run manually to microbenchmark functions, e.g.
"External/dav1d/dav1d_checkasm --bench --test=mc_8bpc --function=warp*".

It is not very meaningful to benchmark the execution of the whole
checkasm executable, as it runs a different numbers of functions
depending on the number of SIMD extensions available on the target
CPU.

(Benchmarking on aarch64 currently requires direct access to
the pmccntr_el0 register. To currently use a different
timer register, edit dav1d/tests/checkasm/checkasm.h and change
pmccntr_el0 into cntvct_el0.)

This uses a static configuration of the project (when building the
upstream project with their own build system, there's a number of
build options that can be configured). Assembly is hooked up and
enabled on i386, x86_64, arm and aarch64.

For architectures other than those, the checkasm test won't have
any reference for detecting e.g. miscompilations of functions,
but building it can still be meaningful (for testing compilation,
or benchmarking the execution of the C version of functions).

Author: mstorsjo

External/CMakeLists.txt

@@ -5,6 +5,7 @@ add_subdirectory(HeCBench)
 add_subdirectory(Nurbs)
 add_subdirectory(Povray)
 add_subdirectory(SPEC)
+add_subdirectory(dav1d)
 add_subdirectory(skidmarks10)
 add_subdirectory(sollve_vv)
 add_subdirectory(smoke)

External/dav1d/CMakeLists.txt

@@ -0,0 +1,368 @@
+include(External)
+
+# git clone -b 1.5.0 https://code.videolan.org/videolan/dav1d.git
+# in llvm-test-suite/test-suite-externals.
+
+llvm_externals_find(TEST_SUITE_DAV1D_ROOT "dav1d" "dav1d 1.5.0")
+
+if (NOT TEST_SUITE_DAV1D_ROOT)
+  return()
+endif()
+
+include(CheckCCompilerFlag)
+include(CheckFunctionExists)
+include(CheckLanguage)
+include(CheckLibraryExists)
+include(CheckLinkerFlag)
+
+set(CMAKE_C_STANDARD 17)
+
+include_directories(.)
+include_directories(${TEST_SUITE_DAV1D_ROOT}/include)
+include_directories(${TEST_SUITE_DAV1D_ROOT}/include/dav1d)
+include_directories(${TEST_SUITE_DAV1D_ROOT})
+include_directories(${TEST_SUITE_DAV1D_ROOT}/src)
+
+if (WIN32)
+  include_directories(${TEST_SUITE_DAV1D_ROOT}/include/compat)
+endif()
+
+function(check_enable_option option)
+  string(REGEX REPLACE "^--*" "" option_name ${option})
+  string(TOUPPER ${option_name} option_name)
+  string(REGEX REPLACE "[-=]" "_" option_name ${option_name})
+  set(varname "SUPPORTS_${option_name}")
+  check_c_compiler_flag(${option} ${varname})
+  if (${varname})
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} ${option}" PARENT_SCOPE)
+  endif()
+endfunction()
+
+function(check_disable_warning warning)
+  string(REGEX REPLACE "^-Wno-" "" warning ${warning})
+  string(TOUPPER ${warning} warning_name)
+  string(REGEX REPLACE "[-=]" "_" warning_name ${warning_name})
+  set(varname "SUPPORTS_W${warning_name}")
+  check_c_compiler_flag(-W${warning} ${varname})
+  if (${varname})
+    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Wno-${warning}" PARENT_SCOPE)
+  endif()
+endfunction()
+
+if (NOT MSVC)
+  # clang-cl supports -Wall, but it corresponds to -Weverything
+  check_enable_option(-Wall)
+endif()
+
+check_enable_option(-Wundef)
+check_enable_option(-Werror=vla)
+check_disable_warning(-Wno-maybe-uninitialized)
+check_disable_warning(-Wno-missing-field-initializers)
+check_disable_warning(-Wno-unused-parameter)
+check_enable_option(-Wstrict-prototypes)
+check_enable_option(-Werror=missing-prototypes)
+check_enable_option(-Wshorten-64-to-32)
+
+check_function_exists(sin HAVE_DEFAULT_MATH)
+if (NOT HAVE_DEFAULT_MATH)
+  check_library_exists(m sin "" HAVE_LIBM)
+  if (HAVE_LIBM)
+    link_libraries(m)
+  endif()
+endif()
+check_library_exists(atomic __atomic_load_8 "" HAVE_LIBATOMIC)
+if (HAVE_LIBATOMIC)
+  link_libraries(atomic)
+endif()
+if (NOT WIN32)
+  find_package(Threads)
+  if (Threads_FOUND)
+    link_libraries(${CMAKE_THREAD_LIBS_INIT})
+  endif()
+endif()
+
+if (WIN32)
+  add_compile_definitions(WIN32_LEAN_AND_MEAN)
+  if (MSVC)
+    add_compile_definitions(_CRT_SECURE_NO_WARNINGS)
+    add_compile_definitions(_CRT_NONSTDC_NO_DEPRECATE)
+  endif()
+endif()
+
+if (CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64" OR CMAKE_SYSTEM_PROCESSOR STREQUAL "arm64")
+  enable_language(ASM)
+  set(CMAKE_SYSTEM_PROCESSOR "aarch64")
+  message(STATUS "dav1d: Enabling aarch64 assembly")
+elseif (CMAKE_SYSTEM_PROCESSOR MATCHES "^arm")
+  enable_language(ASM)
+  set(CMAKE_SYSTEM_PROCESSOR "arm")
+  message(STATUS "dav1d: Enabling arm assembly")
+elseif (CMAKE_SYSTEM_PROCESSOR MATCHES "^i.86$" OR CMAKE_SYSTEM_PROCESSOR MATCHES "^[Xx]86$")
+  check_language(ASM_NASM)
+  if (CMAKE_ASM_NASM_COMPILER)
+    enable_language(ASM_NASM)
+    message(STATUS "dav1d: Enabling i386 nasm assembly")
+  else()
+    add_compile_definitions(NO_X86ASM)
+    message(STATUS "dav1d: Not enabling i386 nasm assembly")
+  endif()
+  set(CMAKE_SYSTEM_PROCESSOR "i386")
+  if (CMAKE_SYSTEM_NAME STREQUAL "Darwin" OR CMAKE_SYSTEM_NAME STREQUAL "Windows")
+    set(CMAKE_ASM_NASM_FLAGS "${CMAKE_ASM_NASM_FLAGS} -DPREFIX")
+  endif()
+elseif (CMAKE_SYSTEM_PROCESSOR STREQUAL "x86_64" OR CMAKE_SYSTEM_PROCESSOR STREQUAL "AMD64" OR CMAKE_SYSTEM_PROCESSOR STREQUAL "amd64")
+  add_compile_definitions(PIC)
+  check_language(ASM_NASM)
+  if (CMAKE_ASM_NASM_COMPILER)
+    enable_language(ASM_NASM)
+    message(STATUS "dav1d: Enabling x86_64 nasm assembly")
+  else()
+    add_compile_definitions(NO_X86ASM)
+    message(STATUS "dav1d: Not enabling x86_64 nasm assembly")
+  endif()
+  set(CMAKE_SYSTEM_PROCESSOR "x86_64")
+  set(CMAKE_ASM_NASM_FLAGS "${CMAKE_ASM_NASM_FLAGS} -DARCH_X86_64=1")
+  if (CMAKE_SYSTEM_NAME STREQUAL "Darwin")
+    set(CMAKE_ASM_NASM_FLAGS "${CMAKE_ASM_NASM_FLAGS} -DPREFIX")
+  endif()
+else()
+  message(STATUS "Not enabling any assembly optimizations for ${CMAKE_SYSTEM_PROCESSOR}")
+endif()
+
+# src
+
+set(dav1d_src
+    cdf.c
+    cpu.c
+    ctx.c
+    data.c
+    decode.c
+    dequant_tables.c
+    getbits.c
+    intra_edge.c
+    itx_1d.c
+    lf_mask.c
+    lib.c
+    log.c
+    mem.c
+    msac.c
+    obu.c
+    pal.c
+    picture.c
+    qm.c
+    ref.c
+    refmvs.c
+    scan.c
+    tables.c
+    thread_task.c
+    warpmv.c
+    wedge.c)
+
+if (WIN32)
+  list(APPEND dav1d_src
+      win32/thread.c)
+endif()
+
+set(dav1d_tmpl_src
+    cdef_apply_tmpl.c
+    cdef_tmpl.c
+    fg_apply_tmpl.c
+    filmgrain_tmpl.c
+    ipred_prepare_tmpl.c
+    ipred_tmpl.c
+    itx_tmpl.c
+    lf_apply_tmpl.c
+    loopfilter_tmpl.c
+    looprestoration_tmpl.c
+    lr_apply_tmpl.c
+    mc_tmpl.c
+    recon_tmpl.c)
+
+if (CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64")
+  list(APPEND dav1d_src
+      arm/cpu.c
+      arm/64/itx.S
+      arm/64/looprestoration_common.S
+      arm/64/msac.S
+      arm/64/refmvs.S
+      arm/64/cdef.S
+      arm/64/filmgrain.S
+      arm/64/ipred.S
+      arm/64/loopfilter.S
+      arm/64/looprestoration.S
+      arm/64/mc.S
+      arm/64/mc_dotprod.S
+      arm/64/cdef16.S
+      arm/64/filmgrain16.S
+      arm/64/ipred16.S
+      arm/64/itx16.S
+      arm/64/loopfilter16.S
+      arm/64/looprestoration16.S
+      arm/64/mc16.S
+      arm/64/mc16_sve.S)
+elseif (CMAKE_SYSTEM_PROCESSOR STREQUAL "arm")
+  list(APPEND dav1d_src
+      arm/cpu.c
+      arm/32/itx.S
+      arm/32/looprestoration_common.S
+      arm/32/msac.S
+      arm/32/refmvs.S
+      arm/32/cdef.S
+      arm/32/filmgrain.S
+      arm/32/ipred.S
+      arm/32/loopfilter.S
+      arm/32/looprestoration.S
+      arm/32/mc.S
+      arm/32/cdef16.S
+      arm/32/filmgrain16.S
+      arm/32/ipred16.S
+      arm/32/itx16.S
+      arm/32/loopfilter16.S
+      arm/32/looprestoration16.S
+      arm/32/mc16.S)
+elseif (CMAKE_SYSTEM_PROCESSOR STREQUAL "x86_64" OR CMAKE_SYSTEM_PROCESSOR STREQUAL "i386")
+  list(APPEND dav1d_src
+      x86/cpu.c)
+  if (CMAKE_ASM_NASM_COMPILER)
+    set(x86_nasm_sources
+        x86/cpuid.asm
+        x86/msac.asm
+        x86/pal.asm
+        x86/refmvs.asm
+        x86/itx_avx512.asm
+        x86/cdef_avx2.asm
+        x86/itx_avx2.asm
+        x86/cdef_sse.asm
+        x86/itx_sse.asm
+        x86/cdef_avx512.asm
+        x86/filmgrain_avx512.asm
+        x86/ipred_avx512.asm
+        x86/loopfilter_avx512.asm
+        x86/looprestoration_avx512.asm
+        x86/mc_avx512.asm
+        x86/filmgrain_avx2.asm
+        x86/ipred_avx2.asm
+        x86/loopfilter_avx2.asm
+        x86/looprestoration_avx2.asm
+        x86/mc_avx2.asm
+        x86/filmgrain_sse.asm
+        x86/ipred_sse.asm
+        x86/loopfilter_sse.asm
+        x86/looprestoration_sse.asm
+        x86/mc_sse.asm
+        x86/cdef16_avx512.asm
+        x86/filmgrain16_avx512.asm
+        x86/ipred16_avx512.asm
+        x86/itx16_avx512.asm
+        x86/loopfilter16_avx512.asm
+        x86/looprestoration16_avx512.asm
+        x86/mc16_avx512.asm
+        x86/cdef16_avx2.asm
+        x86/filmgrain16_avx2.asm
+        x86/ipred16_avx2.asm
+        x86/itx16_avx2.asm
+        x86/loopfilter16_avx2.asm
+        x86/looprestoration16_avx2.asm
+        x86/mc16_avx2.asm
+        x86/cdef16_sse.asm
+        x86/filmgrain16_sse.asm
+        x86/ipred16_sse.asm
+        x86/itx16_sse.asm
+        x86/loopfilter16_sse.asm
+        x86/looprestoration16_sse.asm
+        x86/mc16_sse.asm)
+    list(APPEND dav1d_src
+        ${x86_nasm_sources})
+    list(TRANSFORM x86_nasm_sources PREPEND ${TEST_SUITE_DAV1D_ROOT}/src/)
+    set_source_files_properties(${x86_nasm_sources} PROPERTIES LANGUAGE ASM_NASM)
+  endif()
+endif()
+
+list(TRANSFORM dav1d_tmpl_src PREPEND ${TEST_SUITE_DAV1D_ROOT}/src/)
+list(TRANSFORM dav1d_src PREPEND ${TEST_SUITE_DAV1D_ROOT}/src/)
+
+foreach(bitdepth 8 16)
+  llvm_test_library(dav1d_bitdepth_${bitdepth} OBJECT ${dav1d_tmpl_src})
+  target_compile_definitions(dav1d_bitdepth_${bitdepth} PRIVATE -DBITDEPTH=${bitdepth})
+  list(APPEND bitdepth_libraries dav1d_bitdepth_${bitdepth})
+endforeach()
+
+llvm_test_library(dav1d_lib ${dav1d_src})
+target_link_libraries(dav1d_lib LINK_PRIVATE ${bitdepth_libraries})
+
+
+# tools
+
+set(dav1d_cli_src
+    dav1d.c
+    dav1d_cli_parse.c
+    input/input.c
+    input/annexb.c
+    input/ivf.c
+    input/section5.c
+    output/md5.c
+    output/null.c
+    output/output.c
+    output/y4m2.c
+    output/yuv.c)
+
+if (WIN32)
+  list(APPEND dav1d_cli_src
+      compat/getopt.c)
+endif()
+
+list(TRANSFORM dav1d_cli_src PREPEND ${TEST_SUITE_DAV1D_ROOT}/tools/)
+
+llvm_test_executable_no_test(dav1d ${dav1d_cli_src})
+
+target_include_directories(dav1d PRIVATE ${TEST_SUITE_DAV1D_ROOT}/tools)
+target_link_libraries(dav1d PRIVATE dav1d_lib)
+
+
+# checkasm
+
+set(checkasm_src
+    checkasm.c
+    msac.c
+    pal.c
+    refmvs.c)
+
+set(checkasm_tmpl_src
+    cdef.c
+    filmgrain.c
+    ipred.c
+    itx.c
+    loopfilter.c
+    looprestoration.c
+    mc.c)
+
+if (CMAKE_SYSTEM_PROCESSOR STREQUAL "aarch64")
+  list(APPEND checkasm_src
+      arm/checkasm_64.S)
+elseif (CMAKE_SYSTEM_PROCESSOR STREQUAL "arm")
+  list(APPEND checkasm_src
+      arm/checkasm_32.S)
+elseif (CMAKE_SYSTEM_PROCESSOR STREQUAL "x86_64" OR CMAKE_SYSTEM_PROCESSOR STREQUAL "i386")
+  if (CMAKE_ASM_NASM_COMPILER)
+    set(x86_nasm_sources
+        x86/checkasm.asm)
+    list(APPEND checkasm_src
+        ${x86_nasm_sources})
+    list(TRANSFORM x86_nasm_sources PREPEND ${TEST_SUITE_DAV1D_ROOT}/tests/checkasm/)
+    set_source_files_properties(${x86_nasm_sources} PROPERTIES LANGUAGE ASM_NASM)
+  endif()
+endif()
+
+list(TRANSFORM checkasm_tmpl_src PREPEND ${TEST_SUITE_DAV1D_ROOT}/tests/checkasm/)
+list(TRANSFORM checkasm_src PREPEND ${TEST_SUITE_DAV1D_ROOT}/tests/checkasm/)
+
+foreach(bitdepth 8 16)
+  llvm_test_library(checkasm_bitdepth_${bitdepth} OBJECT ${checkasm_tmpl_src})
+  target_compile_definitions(checkasm_bitdepth_${bitdepth} PRIVATE -DBITDEPTH=${bitdepth})
+  list(APPEND bitdepth_libraries checkasm_bitdepth_${bitdepth})
+endforeach()
+
+llvm_test_run()
+llvm_test_executable(dav1d_checkasm ${checkasm_src})
+target_link_libraries(dav1d_checkasm LINK_PRIVATE ${bitdepth_libraries})
+target_link_libraries(dav1d_checkasm PRIVATE dav1d_lib)

External/dav1d/cli_config.h

@@ -0,0 +1,3 @@
+#pragma once
+
+#define HAVE_XXHASH_H 0

External/dav1d/config.asm

@@ -0,0 +1,11 @@
+%ifdef ARCH_X86_64
+%define ARCH_X86_32 0
+%define STACK_ALIGNMENT 16
+%else
+%define ARCH_X86_32 1
+%define ARCH_X86_64 0
+%define STACK_ALIGNMENT 4
+%endif
+%define FORCE_VEX_ENCODING 0
+%define PIC 1
+%define private_prefix dav1d