image_pack 0.2.0

data/ext/image_pack/extconf.rb

@@ -0,0 +1,515 @@
+# frozen_string_literal: true
+
+require "mkmf"
+require "fileutils"
+require "rbconfig"
+
+if Gem::Version.new(RUBY_VERSION) < Gem::Version.new("3.4.0")
+  abort "image_pack requires Ruby >= 3.4.0; got #{RUBY_VERSION}"
+end
+
+MOZJPEG_CORE_SOURCES = %w[
+  jcapimin.c
+  jcapistd.c
+  jccoefct.c
+  jccolor.c
+  jcdctmgr.c
+  jchuff.c
+  jcext.c
+  jcicc.c
+  jcinit.c
+  jcmainct.c
+  jcmarker.c
+  jcmaster.c
+  jcomapi.c
+  jcparam.c
+  jcphuff.c
+  jcprepct.c
+  jcsample.c
+  jctrans.c
+  jdapimin.c
+  jdapistd.c
+  jdatadst.c
+  jdatasrc.c
+  jdcoefct.c
+  jdcolor.c
+  jddctmgr.c
+  jdhuff.c
+  jdicc.c
+  jdinput.c
+  jdmainct.c
+  jdmarker.c
+  jdmaster.c
+  jdmerge.c
+  jdphuff.c
+  jdpostct.c
+  jdsample.c
+  jdtrans.c
+  jerror.c
+  jfdctflt.c
+  jfdctfst.c
+  jfdctint.c
+  jidctflt.c
+  jidctfst.c
+  jidctint.c
+  jidctred.c
+  jquant1.c
+  jquant2.c
+  jutils.c
+  jmemmgr.c
+  jmemnobs.c
+].freeze
+
+NEON_AARCH64_SOURCES = [
+  "simd/arm/aarch64/jsimd.c",
+  "simd/arm/aarch64/jchuff-neon.c",
+  "simd/arm/jccolor-neon.c",
+  "simd/arm/jcgray-neon.c",
+  "simd/arm/jcphuff-neon.c",
+  "simd/arm/jcsample-neon.c",
+  "simd/arm/jdcolor-neon.c",
+  "simd/arm/jdmerge-neon.c",
+  "simd/arm/jdsample-neon.c",
+  "simd/arm/jfdctfst-neon.c",
+  "simd/arm/jfdctint-neon.c",
+  "simd/arm/jidctfst-neon.c",
+  "simd/arm/jidctint-neon.c",
+  "simd/arm/jidctred-neon.c",
+  "simd/arm/jquanti-neon.c",
+].freeze
+
+X86_64_C_SOURCES = [
+  "simd/x86_64/jsimd.c",
+].freeze
+
+X86_64_ASM_SOURCES = %w[
+  simd/x86_64/jsimdcpu.asm
+  simd/x86_64/jfdctflt-sse.asm
+  simd/x86_64/jccolor-sse2.asm
+  simd/x86_64/jcgray-sse2.asm
+  simd/x86_64/jchuff-sse2.asm
+  simd/x86_64/jcphuff-sse2.asm
+  simd/x86_64/jcsample-sse2.asm
+  simd/x86_64/jdcolor-sse2.asm
+  simd/x86_64/jdmerge-sse2.asm
+  simd/x86_64/jdsample-sse2.asm
+  simd/x86_64/jfdctfst-sse2.asm
+  simd/x86_64/jfdctint-sse2.asm
+  simd/x86_64/jidctflt-sse2.asm
+  simd/x86_64/jidctfst-sse2.asm
+  simd/x86_64/jidctint-sse2.asm
+  simd/x86_64/jidctred-sse2.asm
+  simd/x86_64/jquantf-sse2.asm
+  simd/x86_64/jquanti-sse2.asm
+  simd/x86_64/jccolor-avx2.asm
+  simd/x86_64/jcgray-avx2.asm
+  simd/x86_64/jcsample-avx2.asm
+  simd/x86_64/jdcolor-avx2.asm
+  simd/x86_64/jdmerge-avx2.asm
+  simd/x86_64/jdsample-avx2.asm
+  simd/x86_64/jfdctint-avx2.asm
+  simd/x86_64/jidctint-avx2.asm
+  simd/x86_64/jquanti-avx2.asm
+].freeze
+
+X86_64_ASM_TEMPLATE_SOURCES = %w[
+  simd/x86_64/jccolext-sse2.asm
+  simd/x86_64/jcgryext-sse2.asm
+  simd/x86_64/jdcolext-sse2.asm
+  simd/x86_64/jdmrgext-sse2.asm
+  simd/x86_64/jccolext-avx2.asm
+  simd/x86_64/jcgryext-avx2.asm
+  simd/x86_64/jdcolext-avx2.asm
+  simd/x86_64/jdmrgext-avx2.asm
+].freeze
+
+MOZJPEG_VERSION = "4.1.5"
+MOZJPEG_VERSION_NUMBER = "4001005"
+
+def find_vendor_dir
+  candidates = [
+    File.join(__dir__, "vendor"),
+    File.join(__dir__, "..", "..", "..", "..", "ext", "image_pack", "vendor"),
+    File.expand_path("../../ext/image_pack/vendor", __dir__),
+    File.join(Dir.pwd, "ext", "image_pack", "vendor"),
+  ]
+
+  dir = __dir__
+  6.times do
+    candidates << File.join(dir, "ext", "image_pack", "vendor")
+    dir = File.dirname(dir)
+  end
+
+  candidates.find { |path| File.exist?(File.join(path, ".vendored")) }
+            &.then { |path| File.expand_path(path) }
+end
+
+def find_image_pack_c_dir
+  candidates = [
+    __dir__,
+    File.join(__dir__, "..", "..", "..", "..", "ext", "image_pack"),
+  ]
+
+  dir = __dir__
+  6.times do
+    candidates << File.join(dir, "ext", "image_pack")
+    dir = File.dirname(dir)
+  end
+
+  candidates.find { |path| File.exist?(File.join(path, "image_pack.c")) }
+            &.then { |path| File.expand_path(path) } || __dir__
+end
+
+def msvc?
+  /mswin|mingw|cygwin/i.match?(RUBY_PLATFORM)
+end
+
+def sizeof_size_t
+  value = RbConfig::CONFIG["SIZEOF_SIZE_T"] || RbConfig::CONFIG["SIZEOF_VOIDP"] || RbConfig::CONFIG["SIZEOF_LONG"]
+  value.to_i.positive? ? value.to_i : 8
+end
+
+def inline_keyword
+  msvc? ? "__inline" : "inline __attribute__((always_inline))"
+end
+
+def thread_local_keyword
+  msvc? ? "__declspec(thread)" : "__thread"
+end
+
+def detect_simd_arch
+  case RUBY_PLATFORM
+  when /\A(?:arm64|aarch64)/ then :neon_aarch64
+  when /\A(?:x86_64|x64)/    then :x86_64
+  else                            :none
+  end
+end
+
+def find_nasm
+  ENV["AS_NASM"].presence_or_nil ||
+    %w[nasm yasm].find { |bin| system("which #{bin} >/dev/null 2>&1") }
+end
+
+def write_mozjpeg_config_headers!(build_dir, with_simd:)
+  with_simd_flag = with_simd ? "#define WITH_SIMD 1" : "#undef WITH_SIMD"
+
+  File.write(File.join(build_dir, "jconfig.h"), <<~HEADER)
+    /* Generated by image_pack extconf.rb. */
+    #define JPEG_LIB_VERSION 62
+    #define LIBJPEG_TURBO_VERSION #{MOZJPEG_VERSION}
+    #define LIBJPEG_TURBO_VERSION_NUMBER #{MOZJPEG_VERSION_NUMBER}
+    #undef C_ARITH_CODING_SUPPORTED
+    #undef D_ARITH_CODING_SUPPORTED
+    #define MEM_SRCDST_SUPPORTED 1
+    #{with_simd_flag}
+    #define BITS_IN_JSAMPLE 8
+    #undef RIGHT_SHIFT_IS_UNSIGNED
+  HEADER
+
+  File.write(File.join(build_dir, "jconfigint.h"), <<~HEADER)
+    /* Generated by image_pack extconf.rb. */
+    #define BUILD "image_pack"
+    #undef inline
+    #define INLINE #{inline_keyword}
+    #define THREAD_LOCAL #{thread_local_keyword}
+    #define PACKAGE_NAME "mozjpeg"
+    #define VERSION "#{MOZJPEG_VERSION}"
+    #define SIZEOF_SIZE_T #{sizeof_size_t}
+    #if defined(__has_attribute)
+    #if __has_attribute(fallthrough)
+    #define FALLTHROUGH __attribute__((fallthrough));
+    #else
+    #define FALLTHROUGH
+    #endif
+    #else
+    #define FALLTHROUGH
+    #endif
+  HEADER
+
+  File.write(File.join(build_dir, "jversion.h"), <<~HEADER)
+    /* Generated by image_pack extconf.rb. */
+    #if JPEG_LIB_VERSION >= 80
+    #define JVERSION "8d  15-Jan-2012"
+    #elif JPEG_LIB_VERSION >= 70
+    #define JVERSION "7  27-Jun-2009"
+    #else
+    #define JVERSION "6b  27-Mar-1998"
+    #endif
+
+    #define JCOPYRIGHT \
+      "Copyright (C) 2009-2023 D. R. Commander\\n" \
+      "Copyright (C) 2015, 2020 Google, Inc.\\n" \
+      "Copyright (C) 2019-2020 Arm Limited\\n" \
+      "Copyright (C) 2015-2016, 2018 Matthieu Darbois\\n" \
+      "Copyright (C) 2011-2016 Siarhei Siamashka\\n" \
+      "Copyright (C) 2015 Intel Corporation\\n" \
+      "Copyright (C) 2013-2014 Linaro Limited\\n" \
+      "Copyright (C) 2013-2014 MIPS Technologies, Inc.\\n" \
+      "Copyright (C) 2009, 2012 Pierre Ossman for Cendio AB\\n" \
+      "Copyright (C) 2009-2011 Nokia Corporation and/or its subsidiary(-ies)\\n" \
+      "Copyright (C) 1999-2006 MIYASAKA Masaru\\n" \
+      "Copyright (C) 1991-2020 Thomas G. Lane, Guido Vollbeding"
+
+    #define JCOPYRIGHT_SHORT \
+      "Copyright (C) 1991-2023 The libjpeg-turbo Project and many others"
+  HEADER
+end
+
+def select_simd_backend(mozjpeg_dir, arch)
+  case arch
+  when :neon_aarch64
+    template_in = File.join(mozjpeg_dir, "simd", "arm", "neon-compat.h.in")
+    sources_present = NEON_AARCH64_SOURCES.all? { |rel| File.exist?(File.join(mozjpeg_dir, rel)) }
+    template_present = File.exist?(template_in)
+
+    if sources_present && template_present
+      write_neon_compat_header!(mozjpeg_dir)
+
+      {
+        kind:           :neon_aarch64,
+        c_sources:      NEON_AARCH64_SOURCES,
+        asm_sources:    [],
+        extra_includes: [
+          File.join(mozjpeg_dir, "simd", "arm"),
+          File.join(mozjpeg_dir, "simd", "arm", "aarch64"),
+        ],
+        extra_cflags:   "-DNEON_INTRINSICS=1"
+      }
+    else
+      missing = NEON_AARCH64_SOURCES.reject { |rel| File.exist?(File.join(mozjpeg_dir, rel)) }
+      missing << "simd/arm/neon-compat.h.in" unless template_present
+      warn "image_pack: ARM64 detected but NEON sources missing under #{mozjpeg_dir}:"
+      missing.each { |m| warn "  - #{m}" }
+      warn "image_pack: re-run `bundle exec rake vendor` after updating script/vendor_libs.rb."
+      warn "image_pack: falling back to scalar."
+      { kind: :none }
+    end
+
+  when :x86_64
+    if find_nasm.nil?
+      warn "image_pack: x86_64 detected but NASM/YASM not found on PATH. Falling back to scalar; install nasm for ~3-4x speedup."
+      { kind: :none }
+    elsif !X86_64_C_SOURCES.all? { |rel| File.exist?(File.join(mozjpeg_dir, rel)) }
+      warn "image_pack: x86_64 SIMD sources missing under #{mozjpeg_dir}/simd/x86_64/. Falling back to scalar."
+      { kind: :none }
+    else
+      {
+        kind:           :x86_64_simd,
+        c_sources:      X86_64_C_SOURCES,
+        asm_sources:    X86_64_ASM_SOURCES.select { |rel| File.exist?(File.join(mozjpeg_dir, rel)) },
+        extra_includes: [
+          File.join(mozjpeg_dir, "simd"),
+          File.join(mozjpeg_dir, "simd", "nasm"),
+          File.join(mozjpeg_dir, "simd", "x86_64"),
+        ],
+        extra_cflags:   "",
+      }
+    end
+
+  else
+    { kind: :none }
+  end
+end
+
+NEON_COMPAT_HEADER = <<~HEADER
+  /*
+   * Generated by image_pack extconf.rb (replaces simd/arm/neon-compat.h.in
+   * which CMake would otherwise process). All three intrinsic feature flags
+   * are unconditionally enabled — every AArch64 toolchain image_pack supports
+   * provides them.
+   */
+  #ifndef IMAGE_PACK_NEON_COMPAT_H
+  #define IMAGE_PACK_NEON_COMPAT_H
+
+  #define HAVE_VLD1_S16_X3
+  #define HAVE_VLD1_U16_X2
+  #define HAVE_VLD1Q_U8_X4
+
+  #if defined(_MSC_VER) && !defined(__clang__)
+  #define BUILTIN_CLZ(x)  _CountLeadingZeros(x)
+  #define BUILTIN_CLZLL(x) _CountLeadingZeros64(x)
+  #define BUILTIN_BSWAP64(x) _byteswap_uint64(x)
+  #elif defined(__clang__) || defined(__GNUC__)
+  #define BUILTIN_CLZ(x)  __builtin_clz(x)
+  #define BUILTIN_CLZLL(x) __builtin_clzll(x)
+  #define BUILTIN_BSWAP64(x) __builtin_bswap64(x)
+  #else
+  #error "Unknown compiler"
+  #endif
+
+  #endif /* IMAGE_PACK_NEON_COMPAT_H */
+HEADER
+
+def write_neon_compat_header!(mozjpeg_dir)
+  target = File.join(mozjpeg_dir, "simd", "arm", "neon-compat.h")
+  return target if File.exist?(target) &&
+                   File.read(target) == NEON_COMPAT_HEADER
+  FileUtils.mkdir_p(File.dirname(target))
+  File.write(target, NEON_COMPAT_HEADER)
+  puts "  Generated #{target}"
+  target
+end
+
+class String
+  unless instance_methods.include?(:presence_or_nil)
+    def presence_or_nil
+      empty? ? nil : self
+    end
+  end
+end
+class NilClass
+  unless instance_methods.include?(:presence_or_nil)
+    def presence_or_nil
+      nil
+    end
+  end
+end
+
+def configure_vendored_mozjpeg(vendor_dir)
+  versions = File.read(File.join(vendor_dir, ".vendored"))
+  mozjpeg_dir = File.join(vendor_dir, "mozjpeg")
+
+  abort "vendored MozJPEG sources not found in #{mozjpeg_dir}" unless Dir.exist?(mozjpeg_dir)
+
+  missing_sources = MOZJPEG_CORE_SOURCES.reject { |source| File.exist?(File.join(mozjpeg_dir, source)) }
+  abort "vendored MozJPEG sources are incomplete: #{missing_sources.join(', ')}" unless missing_sources.empty?
+
+  arch = detect_simd_arch
+  backend = select_simd_backend(mozjpeg_dir, arch)
+  with_simd = backend[:kind] != :none
+  simd_c_sources = backend.fetch(:c_sources, [])
+  simd_asm_sources = backend.fetch(:asm_sources, [])
+  scalar_fallback = with_simd ? [] : ["jsimd_none.c"]
+
+  all_c_sources = MOZJPEG_CORE_SOURCES + scalar_fallback + simd_c_sources
+
+  puts "Building with VENDORED MozJPEG from #{mozjpeg_dir}"
+  puts "  #{versions.tr("\n", ", ")}"
+  puts "  arch=#{arch} backend=#{backend[:kind]}"
+  puts "  #{MOZJPEG_CORE_SOURCES.length} core C files + " \
+       "#{simd_c_sources.length} SIMD C files + " \
+       "#{simd_asm_sources.length} ASM files" \
+       "#{with_simd ? '' : ' + jsimd_none.c stub'}"
+
+  build_dir = Dir.pwd
+  write_mozjpeg_config_headers!(build_dir, with_simd: with_simd)
+
+  $CPPFLAGS += " -I#{build_dir}"
+  $CPPFLAGS += " -I#{mozjpeg_dir}"
+
+  vpath_dirs = [find_image_pack_c_dir, mozjpeg_dir]
+
+  if with_simd
+    backend[:extra_includes].each { |inc| $CPPFLAGS += " -I#{inc}" }
+    extra_cflags = backend[:extra_cflags].to_s
+    $CFLAGS += " #{extra_cflags}" unless extra_cflags.empty?
+    simd_c_dirs = simd_c_sources.map { |rel| File.dirname(File.join(mozjpeg_dir, rel)) }.uniq
+    vpath_dirs += simd_c_dirs
+  end
+
+  $srcs = ["image_pack.c"] + all_c_sources
+  $VPATH = vpath_dirs.uniq
+
+  {
+    backend:       backend,
+    with_simd:     with_simd,
+    vpath_dirs:    vpath_dirs.uniq,
+    asm_sources:   simd_asm_sources,
+    mozjpeg_dir:   mozjpeg_dir,
+  }
+end
+
+def patch_makefile_vpath!(vpath_dirs)
+  makefile = File.read("Makefile")
+  return if makefile.include?("# vendored mozjpeg vpath")
+
+  vpath_lines = vpath_dirs.map { |dir| "vpath %.c #{dir}" }.join("\n")
+
+  makefile.sub!(/^(VPATH\s*=[^\n]*)$/) { "#{Regexp.last_match(1)}\n# vendored mozjpeg vpath\n#{vpath_lines}" }
+  File.write("Makefile", makefile)
+  puts "  Patched Makefile with #{vpath_dirs.length} VPATH entries"
+end
+
+def patch_makefile_nasm!(asm_sources, mozjpeg_dir)
+  return if asm_sources.empty?
+
+  nasm = find_nasm
+  abort "NASM/YASM disappeared between probe and Makefile patch?" unless nasm
+
+  fmt, platform_define, pic_define =
+    case RUBY_PLATFORM
+    when /darwin/
+      ["macho64", "MACHO", nil]
+    when /linux/
+      ["elf64", "ELF", "PIC"]
+    when /mswin|mingw|cygwin/
+      ["win64", "WIN64", nil]
+    else
+      ["elf64", "ELF", "PIC"]
+    end
+
+  obj_ext = RbConfig::CONFIG["OBJEXT"] || "o"
+
+  asm_objs = asm_sources.map { |rel| File.basename(rel, ".asm") + ".#{obj_ext}" }
+  inc_dirs = [
+    File.join(mozjpeg_dir, "simd"),
+    File.join(mozjpeg_dir, "simd", "nasm"),
+    File.join(mozjpeg_dir, "simd", "x86_64"),
+    File.join(mozjpeg_dir, "win"),
+  ].select { |d| Dir.exist?(d) }
+
+  nasm_includes = inc_dirs.map { |d| "-I#{d}/" }.join(" ")
+  arch_defines = %w[elf64 macho64 win64].include?(fmt) ? ["__x86_64__"] : []
+  nasm_defines = ([platform_define, pic_define] + arch_defines).compact.map { |name| "-D#{name}" }.join(" ")
+
+  rules = +""
+  asm_sources.each do |rel|
+    src = File.join(mozjpeg_dir, rel)
+    obj = File.basename(rel, ".asm") + ".#{obj_ext}"
+    rules << "#{obj}: #{src}\n"
+    rules << "\t#{nasm} -f#{fmt} #{nasm_defines} #{nasm_includes} -o $@ $<\n"
+  end
+
+  makefile = File.read("Makefile")
+  return if makefile.include?("# vendored mozjpeg nasm")
+
+  asm_obj_line = asm_objs.join(" ")
+  makefile.sub!(/^(OBJS\s*=[^\n]*)$/) { "#{Regexp.last_match(1)} #{asm_obj_line}\n# vendored mozjpeg nasm" }
+  makefile << "\n# vendored mozjpeg nasm rules\n"
+  makefile << rules
+  File.write("Makefile", makefile)
+  puts "  Patched Makefile with #{asm_sources.length} NASM rules (#{nasm}, -f#{fmt})"
+end
+
+vendor_dir = find_vendor_dir
+
+unless vendor_dir
+  abort <<~MSG
+    vendored native library is missing.
+    Run `bundle exec rake vendor` before building the gem package.
+    Published gems must include ext/image_pack/vendor/.vendored and ext/image_pack/vendor/mozjpeg/*.
+  MSG
+end
+
+build_info = configure_vendored_mozjpeg(vendor_dir)
+
+$warnflags = ""
+
+unless msvc?
+  $CFLAGS += " -O3 -Wall -Wextra -Wno-unused-parameter -Wno-sign-compare -std=gnu11"
+else
+  $CFLAGS += " -O2"
+end
+
+$CFLAGS += " -DIMAGE_PACK_RUBY34_ONLY=1"
+$CFLAGS += build_info[:with_simd] ? " -DIMAGE_PACK_HAS_SIMD=1" : " -DIMAGE_PACK_PURE_C=1"
+$LIBS += " -lm" unless msvc?
+
+have_header("ruby/thread.h") or abort "ruby/thread.h not found"
+have_header("ruby/version.h") or abort "ruby/version.h not found"
+have_header("jpeglib.h") or abort "jpeglib.h not found in vendored MozJPEG include paths"
+
+create_makefile("image_pack/image_pack")
+
+patch_makefile_vpath!(build_info[:vpath_dirs])
+patch_makefile_nasm!(build_info[:asm_sources], build_info[:mozjpeg_dir])