Port zlib-ng to the Android platform

- Add Soong blueprint
- Add CRC32 fixes
- Add NDK mappings

All changes were based on the Android fork of Chromium zlib.

As reported by simpleperf, a considerable amount of CPU time is spent
decompressing Android binary XML resources when opening and closing
activities in Settings. Because of this, some binary XMLs from
framework-res.apk have been used for zlib_bench testing.

zlib-ng is significantly faster than vanilla zlib and Chromium zlib
(with optimizations enabled) in terms of compression. 64-bit benchmarks
with zlib_bench64 on Android:

Compression:
+------------------------------------------+------+---------------+---------+-------+
|                   File                   | zlib | Chromium zlib | zlib-ng |  Unit |
+------------------------------------------+------+---------------+---------+-------+
|     res/anim/activity_open_enter.xml     | 10.4 |       16      |   25.1  | MiB/s |
|    res/anim/activity_close_enter.xml     | 9.3  |      15.5     |   21.1  | MiB/s |
|         res/transition/move.xml          | 9.7  |      16.4     |   21.6  | MiB/s |
| res/drawable/stat_sys_battery_charge.xml | 13.3 |      22.4     |   33.2  | MiB/s |
+------------------------------------------+------+---------------+---------+-------+

The gains are less significant for decompression, and Chromium zlib is
generally slighty faster in this case. However, zlib-ng still tends to
be faster than vanilla zlib:

Decompression:
+------------------------------------------+-------+---------------+---------+-------+
|                   File                   |  zlib | Chromium zlib | zlib-ng |  Unit |
+------------------------------------------+-------+---------------+---------+-------+
|     res/anim/activity_open_enter.xml     | 136.8 |     150.6     |  153.2  | MiB/s |
|    res/anim/activity_close_enter.xml     | 125.7 |     133.5     |   123   | MiB/s |
|         res/transition/move.xml          | 120.6 |     129.9     |  124.4  | MiB/s |
| res/drawable/stat_sys_battery_charge.xml | 194.6 |     222.1     |  217.3  | MiB/s |
+------------------------------------------+-------+---------------+---------+-------+

In combination with enabling -O3 for libpng, zlib-ng speeds up
screenshot saving by 16% on redfin (Cortex-A55 + Cortex-A76) as compared
to vanilla zlib.

While Android is likely to be decompressing much more data than it
compresses, zlib-ng's significantly faster compression should be a
worthwhile tradeoff for its deficiencies in decompression speed.

zlib versions used for benchmarks:
Vanilla zlib: [1] (Chromium zlib with optimizations disabled)
Chromium zlib: [2] with all optimizations enabled
zlib-ng: [3] with Android porting changes
All zlib variants were built with -O3, as per commit [4].

[1] https://android.googlesource.com/platform/external/zlib/+/refs/tags/android-11.0.0_r35
[2] https://android.googlesource.com/platform/external/zlib/+/e620e8065d17bab7c9eefe3a03d7f8b80ca450d6
[3] https://github.com/zlib-ng/zlib-ng/commit/9f784908599006e9f09599b85b495bdcda1f25c4
[4] https://android.googlesource.com/platform/external/zlib/+/18e812de

4 files changed, 473 insertions, 0 deletions

diff --git a/Android.bp b/Android.bp
new file mode 100644
index 0000000..bc855be
--- /dev/null
+++ b/Android.bp
@@ -0,0 +1,244 @@
+srcs_arm = [
+    "arch/arm/armfeature.c",
+    "arch/arm/crc32_acle.c",
+    "arch/arm/insert_string_acle.c",
+    "arch/arm/adler32_neon.c",
+    "arch/arm/chunkset_neon.c",
+    "arch/arm/slide_neon.c",
+]
+
+// Not all CPUs will support these features, but compatibility is checked at runtime.
+cflags_arm = [
+    "-DARM_ACLE_CRC_HASH",
+    "-DARM_FEATURES",
+    "-DARM_NEON_ADLER32",
+    "-DARM_NEON_CHUNKSET",
+    "-DARM_NEON_SLIDEHASH",
+]
+
+srcs_x86 = [
+    "arch/x86/x86.c",
+    "arch/x86/slide_avx.c",
+    "arch/x86/chunkset_avx.c",
+    "arch/x86/compare258_avx.c",
+    "arch/x86/adler32_avx.c",
+    "arch/x86/insert_string_sse.c",
+    "arch/x86/compare258_sse.c",
+    "arch/x86/chunkset_sse.c",
+    "arch/x86/slide_sse.c",
+    "arch/x86/adler32_ssse3.c",
+    "arch/x86/crc_folding.c",
+]
+
+cflags_x86 = [
+    "-DX86_AVX2",
+    "-DX86_AVX2_ADLER32",
+    "-DX86_AVX_CHUNKSET",
+    "-DX86_FEATURES",
+    "-DX86_PCLMULQDQ_CRC",
+    "-DX86_SSE2",
+    "-DX86_SSE2_CHUNKSET",
+    "-DX86_SSE2_SLIDEHASH",
+    "-DX86_SSE42_CMP_STR",
+    "-DX86_SSE42_CRC_HASH",
+    "-DX86_SSE42_CRC_INTRIN",
+    "-DX86_SSSE3",
+    "-DX86_SSSE3_ADLER32",
+    "-mavx2",
+    "-msse4",
+    "-mssse3",
+    "-mpclmul",
+]
+
+// This optimization is applicable to arm64 and x86-64.
+cflags_64 = ["-DUNALIGNED64_OK"]
+
+libz_srcs = [
+    "adler32.c",
+    "chunkset.c",
+    "compare258.c",
+    "compress.c",
+    "crc32.c",
+    "crc32_comb.c",
+    "deflate.c",
+    "deflate_fast.c",
+    "deflate_medium.c",
+    "deflate_quick.c",
+    "deflate_slow.c",
+    "functable.c",
+    "infback.c",
+    "inffast.c",
+    "inflate.c",
+    "inftrees.c",
+    "insert_string.c",
+    "trees.c",
+    "uncompr.c",
+    "zutil.c",
+    "gzlib.c",
+    "gzread.c",
+    "gzwrite.c",
+]
+
+cflags_common = [
+    "-DHAVE_BUILTIN_CTZ",
+    "-DHAVE_BUILTIN_CTZLL",
+    "-DHAVE_VISIBILITY_HIDDEN",
+    "-DHAVE_VISIBILITY_INTERNAL",
+    "-DZLIB_CONST",
+
+    "-DWITH_GZFILEOP",
+    "-DZLIB_COMPAT",
+    "-DZLIB_DLL",
+    "-D_LARGEFILE64_SOURCE=1",
+    "-D__USE_LARGEFILE64",
+    "-Wall",
+    "-Werror",
+    "-Wno-implicit-fallthrough",
+    "-O3",
+    "-DNDEBUG",
+    "-fno-semantic-interposition",
+    "-std=c99",
+    "-DUNALIGNED_OK",
+]
+
+cc_defaults {
+    name: "libz_defaults",
+
+    cflags: cflags_common,
+    stl: "none",
+    export_include_dirs: ["."],
+    srcs: libz_srcs,
+
+    arch: {
+        arm: {
+            // TODO: This is to work around b/24465209. Remove after root cause
+            // is fixed.
+            pack_relocations: false,
+            ldflags: ["-Wl,--hash-style=both"],
+
+            cflags: cflags_arm,
+            srcs: srcs_arm,
+        },
+        arm64: {
+            cflags: cflags_arm + cflags_64,
+            srcs: srcs_arm,
+        },
+        x86: {
+            cflags: cflags_x86,
+            srcs: srcs_x86,
+        },
+        x86_64: {
+            cflags: cflags_x86 + cflags_64,
+            srcs: srcs_x86,
+        },
+    },
+}
+
+cc_library {
+    name: "libz",
+    defaults: ["libz_defaults"],
+
+    host_supported: true,
+    unique_host_soname: true,
+    static_ndk_lib: true,
+
+    vendor_available: true,
+    vndk: {
+        enabled: true,
+        support_system_process: true,
+    },
+    ramdisk_available: true,
+    recovery_available: true,
+    native_bridge_supported: true,
+
+    target: {
+        linux_bionic: {
+            enabled: true,
+        },
+        windows: {
+            enabled: true,
+        },
+    },
+
+// TODO(b/155456180): make libz a stub-providing library by uncommenting below
+//    stubs: {
+//        versions: ["29", "30"],
+//        symbol_file: "libz.map.txt",
+//    },
+    apex_available: [
+        "//apex_available:platform",
+        "com.android.art.debug", // from libdexfile
+        "com.android.art.release",
+        "com.android.bluetooth.updatable",
+        "com.android.runtime",
+    ],
+}
+
+// A more stable build of libz. Build configuration of this library should be
+// the same for different targets. This is only used by imgdiff.
+
+cc_library {
+    name: "libz_stable",
+    visibility: [
+        "//bootable/recovery/applypatch",
+        "//bootable/recovery/tests",
+    ],
+    cflags: cflags_common,
+    stl: "none",
+    export_include_dirs: ["."],
+    srcs: libz_srcs,
+
+    host_supported: true,
+    vendor_available: true,
+    recovery_available: true,
+}
+
+cc_binary_host {
+    name: "minigzip",
+    srcs: ["test/minigzip.c"],
+    cflags: cflags_common,
+    static_libs: ["libz"],
+    stl: "none",
+}
+
+// This module is defined in development/ndk/Android.bp. Updating these headers
+// to be usable for any API level is going to be some work (at the very least,
+// there's a ZLIB_VERNUM that will need to be handled since early versions of
+// Android did not have all the APIs that calling code will use if this is set
+// to the current value.
+//
+// The NDK never updated the zlib headers when the platform updated, so until we
+// solve this the NDK will continue shipping the old headers.
+//
+// ndk_headers {
+//     name: "libz_headers",
+//     from: "src",
+//     to: "",
+//     srcs: [
+//         "src/zconf.h",
+//         "src/zlib.h",
+//     ],
+//     license: "NOTICE",
+// }
+
+// TODO(b/155351357) remove this library and let libtextclassifier to use libz
+// instead.
+// libz_current allows modules building against the NDK to have access to zlib
+// API that's not available from the NDK libz.
+cc_library_static {
+    name: "libz_current",
+    defaults: ["libz_defaults"],
+    sdk_version: "current",
+
+    apex_available: [
+        "//apex_available:platform", // indirectly from GoogleExtServices that gets installed to /system
+        "com.android.extservices", // indirectly via libtextclassifier
+    ],
+}
+
+ndk_library {
+    name: "libz",
+    symbol_file: "libz.map.txt",
+    first_version: "9",
+    unversioned_until: "current",
+}
diff --git a/arch/arm/crc32_acle.c b/arch/arm/crc32_acle.c
index 88ba6c3..99013e1 100644
--- a/arch/arm/crc32_acle.c
+++ b/arch/arm/crc32_acle.c
@@ -11,6 +11,46 @@
 #endif
 #include "../../zutil.h"
 
+#if defined(__clang__)
+/* CRC32 intrinsics are #ifdef'ed out of arm_acle.h unless we build with an
+ * armv8 target, which is incompatible with ThinLTO optimizations on Android.
+ * (Namely, mixing and matching different module-level targets makes ThinLTO
+ * warn, and Android defaults to armv7-a. This restriction does not apply to
+ * function-level `target`s, however.)
+ *
+ * Since we only need four crc intrinsics, and since clang's implementation of
+ * those are just wrappers around compiler builtins, it's simplest to #define
+ * those builtins directly. If this #define list grows too much (or we depend on
+ * an intrinsic that isn't a trivial wrapper), we may have to find a better way
+ * to go about this.
+ *
+ * NOTE: clang currently complains that "'+soft-float-abi' is not a recognized
+ * feature for this target (ignoring feature)." This appears to be a harmless
+ * bug in clang.
+ */
+#define __crc32b __builtin_arm_crc32b
+#define __crc32d __builtin_arm_crc32d
+#define __crc32w __builtin_arm_crc32w
+#define __crc32cw __builtin_arm_crc32cw
+#define __crc32h __builtin_arm_crc32h
+
+#if defined(__aarch64__)
+#define TARGET_ARMV8_WITH_CRC __attribute__((target("crc")))
+#else  // !defined(__aarch64__)
+#define TARGET_ARMV8_WITH_CRC __attribute__((target("armv8-a,crc")))
+#endif  // defined(__aarch64__)
+
+#elif defined(__GNUC__)
+/* For GCC, we are setting CRC extensions at module level, so ThinLTO is not
+ * allowed. We can just include arm_acle.h.
+ */
+#include <arm_acle.h>
+#define TARGET_ARMV8_WITH_CRC
+#else  // !defined(__GNUC__) && !defined(_aarch64__)
+#error ARM CRC32 SIMD extensions only supported for Clang and GCC
+#endif
+
+TARGET_ARMV8_WITH_CRC
 uint32_t crc32_acle(uint32_t crc, const unsigned char *buf, uint64_t len) {
     Z_REGISTER uint32_t c;
     Z_REGISTER const uint16_t *buf2;
diff --git a/arch/arm/insert_string_acle.c b/arch/arm/insert_string_acle.c
index 2daf9ba..5553252 100644
--- a/arch/arm/insert_string_acle.c
+++ b/arch/arm/insert_string_acle.c
@@ -12,6 +12,44 @@
 #include "../../zbuild.h"
 #include "../../deflate.h"
 
+#if defined(__clang__)
+/* CRC32 intrinsics are #ifdef'ed out of arm_acle.h unless we build with an
+ * armv8 target, which is incompatible with ThinLTO optimizations on Android.
+ * (Namely, mixing and matching different module-level targets makes ThinLTO
+ * warn, and Android defaults to armv7-a. This restriction does not apply to
+ * function-level `target`s, however.)
+ *
+ * Since we only need four crc intrinsics, and since clang's implementation of
+ * those are just wrappers around compiler builtins, it's simplest to #define
+ * those builtins directly. If this #define list grows too much (or we depend on
+ * an intrinsic that isn't a trivial wrapper), we may have to find a better way
+ * to go about this.
+ *
+ * NOTE: clang currently complains that "'+soft-float-abi' is not a recognized
+ * feature for this target (ignoring feature)." This appears to be a harmless
+ * bug in clang.
+ */
+#define __crc32b __builtin_arm_crc32b
+#define __crc32d __builtin_arm_crc32d
+#define __crc32w __builtin_arm_crc32w
+#define __crc32cw __builtin_arm_crc32cw
+
+#if defined(__aarch64__)
+#define TARGET_ARMV8_WITH_CRC __attribute__((target("crc")))
+#else  // !defined(__aarch64__)
+#define TARGET_ARMV8_WITH_CRC __attribute__((target("armv8-a,crc")))
+#endif  // defined(__aarch64__)
+
+#elif defined(__GNUC__)
+/* For GCC, we are setting CRC extensions at module level, so ThinLTO is not
+ * allowed. We can just include arm_acle.h.
+ */
+#include <arm_acle.h>
+#define TARGET_ARMV8_WITH_CRC
+#else  // !defined(__GNUC__) && !defined(_aarch64__)
+#error ARM CRC32 SIMD extensions only supported for Clang and GCC
+#endif
+
 #define UPDATE_HASH(s, h, val) \
     h = __crc32w(0, val)
 
diff --git a/libz.map.txt b/libz.map.txt
new file mode 100644
index 0000000..850bbf8
--- /dev/null
+++ b/libz.map.txt
@@ -0,0 +1,151 @@
+# This file is copied from src/zlib.map and annotated with comments for the NDK
+# stub library generation script.
+ZLIB_1.2.0 {
+  global:
+    compressBound;
+    deflateBound;
+    inflateBack;
+    inflateBackEnd;
+    inflateBackInit_;
+    inflateCopy;
+  local:
+    deflate_copyright; # var
+    inflate_copyright; # var
+    inflate_fast;
+    inflate_table;
+    zcalloc;
+    zcfree;
+    z_errmsg; # var
+    gz_error;
+    gz_intmax;
+    _*;
+};
+
+ZLIB_1.2.0.2 {
+    gzclearerr;
+    gzungetc;
+    zlibCompileFlags;
+} ZLIB_1.2.0;
+
+ZLIB_1.2.0.8 {
+    deflatePrime;
+} ZLIB_1.2.0.2;
+
+ZLIB_1.2.2 {
+    adler32_combine;
+    crc32_combine;
+    deflateSetHeader;
+    inflateGetHeader;
+} ZLIB_1.2.0.8;
+
+ZLIB_1.2.2.3 {
+    deflateTune;
+    gzdirect;
+} ZLIB_1.2.2;
+
+ZLIB_1.2.2.4 {
+    inflatePrime;
+} ZLIB_1.2.2.3;
+
+ZLIB_1.2.3.3 {
+    adler32_combine64;
+    crc32_combine64;
+    gzopen64;
+    gzseek64;
+    gztell64;
+    inflateUndermine;
+} ZLIB_1.2.2.4;
+
+ZLIB_1.2.3.4 {
+    inflateReset2;
+    inflateMark;
+} ZLIB_1.2.3.3;
+
+ZLIB_1.2.3.5 {
+    gzbuffer;
+    gzoffset;
+    gzoffset64;
+    gzclose_r;
+    gzclose_w;
+} ZLIB_1.2.3.4;
+
+ZLIB_1.2.5.1 {
+    deflatePending;
+} ZLIB_1.2.3.5;
+
+ZLIB_1.2.5.2 {
+    deflateResetKeep;
+    gzgetc_;
+    inflateResetKeep;
+} ZLIB_1.2.5.1;
+
+ZLIB_1.2.7.1 { # introduced=19
+    inflateGetDictionary;
+    gzvprintf;
+} ZLIB_1.2.5.2;
+
+ZLIB_1.2.9 { # introduced=28
+    inflateCodesUsed;
+    inflateValidate;
+    uncompress2;
+    gzfread;
+    gzfwrite;
+    deflateGetDictionary;
+    adler32_z;
+    crc32_z;
+} ZLIB_1.2.7.1;
+
+# These were all exposed by the old NDK stub library. Unclear if they still
+# should be, but at least some of them are marked as being exported in zlib.h
+# and the tree doesn't build without them.
+ZLIB_NDK {
+    _dist_code;
+    _length_code;
+    _tr_align;
+    _tr_flush_bits; # introduced=21
+    _tr_flush_block;
+    _tr_init;
+    _tr_stored_block;
+    _tr_tally;
+    adler32;
+    compress2;
+    compress;
+    crc32;
+    deflate;
+    deflateCopy;
+    deflateEnd;
+    deflateInit2_;
+    deflateInit_;
+    deflateParams;
+    deflateReset;
+    deflateSetDictionary;
+    get_crc_table;
+    gzclose;
+    gzdopen;
+    gzeof;
+    gzerror;
+    gzflush;
+    gzgetc;
+    gzgets;
+    gzopen;
+    gzprintf;
+    gzputc;
+    gzputs;
+    gzread;
+    gzrewind;
+    gzseek;
+    gzsetparams;
+    gztell;
+    gzwrite;
+    inflate;
+    inflateEnd;
+    inflateInit2_;
+    inflateInit_;
+    inflateReset;
+    inflateSetDictionary;
+    inflateSync;
+    inflateSyncPoint;
+    uncompress;
+    zError;
+    zlibVersion;
+};