add -flto=thin to mac aarch64 build cflags for ez 10% gain (#469)

Adds somewhere between 5-10% improvement to wall time. I've ran all tests, booted several fakeships and comets with this installed, and all seems to work properly. Mac x86 is coming, I just don't have access to an intel mac right now [but soon...]. Linux will come later once I figure out some bugs. The following is time to boot a fakeship from a brass pill: ``` Without LTO: ________________________________________________________ Executed in 180.46 secs fish external usr time 17.83 secs 0.10 millis 17.83 secs sys time 0.23 secs 1.86 millis 0.23 secs With LTO: ________________________________________________________ Executed in 164.87 secs fish external usr time 15.65 secs 0.12 millis 15.65 secs sys time 0.19 secs 1.94 millis 0.19 secs ``` 180/164 -> 1.09 --- ~barter-simsum ## Brass pill boot: x86 linux without lto: ~140s x86 linux with lto: ~128s ~ 8.6% improvement - note we may be able to squeeze a bit more out if we can apply `-flto` to all dependencies and not just urbit binaries. This is currently an issue for x86 linux though due to some weird uninvestigated behavior with libsigsegv ## For those curious, the following hints at what was inlined ``` readelf -s ./bazel-bin/pkg/vere/urbit | grep lto_priv 1233: 000000000042d0f0 361 FUNC LOCAL DEFAULT 2 _rebalance.lto_priv.1 1237: 000000000042d6c0 435 FUNC LOCAL DEFAULT 2 _rebalance.lto_priv.0 13002: 0000000000d24a00 200 OBJECT GLOBAL HIDDEN 14 u3_Signal.lto_priv.0 13358: 0000000000d36de0 8 OBJECT GLOBAL HIDDEN 14 sec_u.lto_priv.0 13429: 0000000000d24ae0 80 OBJECT GLOBAL HIDDEN 14 u3V.lto_priv.0 13785: 00000000004516e0 110 FUNC GLOBAL HIDDEN 2 _tap_in.lto_priv.0 14094: 0000000000d3aeb0 4 OBJECT GLOBAL HIDDEN 14 sag_w.lto_priv.0 14100: 0000000000469af0 971 FUNC GLOBAL HIDDEN 2 _lord_stop.lto_priv.0 14120: 00000000004581f0 393 FUNC GLOBAL HIDDEN 2 _cj_nail.lto_priv.0 14178: 0000000000456a30 450 FUNC GLOBAL HIDDEN 2 _cj_fine.lto_priv.0 14732: 0000000000463eb0 82 FUNC GLOBAL HIDDEN 2 _box_free.lto_priv.0 14849: 000000000044a320 658 FUNC GLOBAL HIDDEN 2 _n_find.lto_priv.0 15325: 0000000000478dd0 1487 FUNC GLOBAL HIDDEN 2 _pier_init.lto_priv.0 15413: 00000000004674a0 939 FUNC GLOBAL HIDDEN 2 _ca_willoc.lto_priv.0 15841: 0000000000477300 394 FUNC GLOBAL HIDDEN 2 _dawn_fail.lto_priv.0 16128: 00000000004449d0 1746 FUNC GLOBAL HIDDEN 2 _cr_sing.lto_priv.0 16214: 0000000000474c40 500 FUNC GLOBAL HIDDEN 2 _ttyf_loja.lto_priv.0 16277: 000000000078fe48 8 OBJECT GLOBAL HIDDEN 5 ver_hos_c.lto_priv.0 16887: 0000000000443ab0 460 FUNC GLOBAL HIDDEN 2 _n_bam.lto_priv.0 17346: 000000000045cc90 540 FUNC GLOBAL HIDDEN 2 _cj_minx.lto_priv.0 17897: 0000000000d249e8 8 OBJECT GLOBAL HIDDEN 14 _file_u.lto_priv.0 18325: 00000000004447a0 58 FUNC GLOBAL HIDDEN 2 _n_feb.lto_priv.0 18517: 00000000004453e0 3981 FUNC GLOBAL HIDDEN 2 _n_comp.lto_priv.0 18739: 0000000000d249f6 1 OBJECT GLOBAL HIDDEN 14 _ct_lop_o.lto_priv.0 18880: 000000000044fda0 1650 FUNC GLOBAL HIDDEN 2 _find_home.lto_priv.0 19091: 0000000000474e40 500 FUNC GLOBAL HIDDEN 2 _ttyf_hija.lto_priv.0 19508: 0000000000450420 1393 FUNC GLOBAL HIDDEN 2 _pave_home.lto_priv.0 19595: 000000000042a3d0 687 FUNC GLOBAL HIDDEN 2 _in_uni.lto_priv.0 19822: 0000000000430e20 1095 FUNC GLOBAL HIDDEN 2 _block_rip.lto_priv.0 20236: 0000000000457030 331 FUNC GLOBAL HIDDEN 2 _cj_axis.lto_priv.0 20306: 0000000000457180 514 FUNC GLOBAL HIDDEN 2 _cj_gust.lto_priv.0 20866: 0000000000456c00 667 FUNC GLOBAL HIDDEN 2 _cj_cast.lto_priv.0 21317: 0000000000446fe0 11412 FUNC GLOBAL HIDDEN 2 _n_burn.lto_priv.0 21430: 000000000045c820 1130 FUNC GLOBAL HIDDEN 2 _cj_spot.lto_priv.0 ```
2024-09-11 11:55:31 +03:00 · 2023-07-18 11:59:43 -04:00 · 2023-07-18 11:59:43 -04:00 · 50d602ef8b
commit 50d602ef8b
parent e37e1f1a89 6fe540959d
12 changed files with 117 additions and 27 deletions
--- a/.bazelrc
+++ b/.bazelrc
@ -20,19 +20,21 @@ build --@io_bazel_rules_docker//transitions:enable=false
 build --flag_alias=clang_version=//:clang_version
 build --flag_alias=gcc_version=//:gcc_version

+# Use optimized build by default. According to the bazel documentation, this
+# corresponds to -O2 -DNDEBUG, but these are overriden in
+# //bazel/common_settings.bzl:vere_library.
+# https://bazel.build/docs/user-manual#build-semantics
+build --compilation_mode=opt
+
 # Don't include source level debug info on macOS. See
 # https://github.com/urbit/urbit/issues/5561 and
 # https://github.com/urbit/vere/issues/131.
-build:linux --per_file_copt='pkg/.*@-g'
 build:linux --host_copt='-g'
 build --strip=never

-# Use -O3 as the default optimization level.
-build --per_file_copt='pkg/.*@-O3'
+# Turn on optimization, CPU and memory debug for exec config, which we only use
+# to run the fake ship tests.  Also turn on extra snapshot validation.
 build --host_copt='-O3'
-
-# Turn on CPU and memory debug for exec config, which we only use to run the
-# fake ship tests.  Also turn on extra snapshot validation.
 build --host_copt='-DU3_CPU_DEBUG'
 build --host_copt='-DU3_MEMORY_DEBUG'
 build --host_copt='-DC3DBG'
@ -45,12 +47,5 @@ build:mem_dbg --per_file_copt='pkg/.*@-DU3_MEMORY_DEBUG'
 build:cpu_dbg --per_file_copt='pkg/.*@-DU3_CPU_DEBUG'
 build:snp_dbg --per_file_copt='pkg/.*@-DU3_SNAPSHOT_VALIDATION'

-# Enable maximum debug info and disable optimizations for debug config. It's
-# important that these lines come after setting the default debug and
-# optimization level flags above.
-build:dbg --per_file_copt='pkg/.*@-O0'
-build:dbg --per_file_copt='pkg/.*@-g3'
-build:dbg --per_file_copt='pkg/.*@-DC3DBG'
-
 # Any personal configuration should go in .user.bazelrc.
 try-import %workspace%/.user.bazelrc
--- a/BUILD.bazel
+++ b/BUILD.bazel
@ -37,6 +37,38 @@ config_setting(
    ],
 )

+#
+# CONFIGS DETAILING WHEN TO ENABLE CERTAIN FEATURES BY DEFAULT.
+# CHANGES BEHAVIOR OF //bazel/common_settings.bzl:vere_library.
+#
+
+config_setting(
+    name = "thinlto",
+    constraint_values = [
+        "@platforms//os:macos",
+    ],
+    values = {
+        "compilation_mode": "opt"
+    }
+)
+
+config_setting(
+    name = "lto",
+    constraint_values = [
+        "@platforms//os:linux",
+    ],
+    values = {
+        "compilation_mode": "opt"
+    }
+)
+
+config_setting(
+    name = "debug",
+    values = {
+        "compilation_mode": "dbg"
+    }
+)
+
 #
 # COMPILERS
 #
--- a/INSTALL.md
+++ b/INSTALL.md
@ -60,9 +60,10 @@ bazel build :urbit
 ```

 If you want a debug build, which changes the optimization level from `-O3` to
-`-O0` and includes more debugging information, specify the `dbg` configuration:
+`-O0` and includes more debugging information, specify `dbg` as the
+`compilation_mode`:
 ```console
-bazel build --config=dbg :urbit
+bazel build --compilation_mode=dbg :urbit
 ```
 Note that you cannot change the optimization level for third party
 dependencies--those targets specified in `bazel/third_party`--from the command
--- a/bazel/common_settings.bzl
+++ b/bazel/common_settings.bzl
@ -10,3 +10,50 @@ string_flag = rule(
    build_setting = config.string(flag = True),
    doc = "A string-typed build setting that can be set on the command line",
 )
+
+def vere_library(copts = [], linkopts = [], **kwargs):
+  native.cc_library(
+    copts = copts + select({
+        "//:debug": ["-O0", "-g3", "-DC3DBG"],
+        "//conditions:default": ["-O3"]
+    }) + select({
+        "//:lto": ['-flto'],
+        "//:thinlto": ['-flto=thin'],
+        "//conditions:default": []
+    }) + select({
+        # Don't include source level debug info on macOS. See
+        # https://github.com/urbit/urbit/issues/5561 and
+        # https://github.com/urbit/vere/issues/131.
+        "//:debug": [],
+        "@platforms//os:linux": ["-g"],
+        "//conditions:default": [],
+    }),
+    linkopts = linkopts + ['-g'] + select({
+        "//:lto": ['-flto'],
+        "//:thinlto": ['-flto=thin'],
+        "//conditions:default": []
+    }),
+    **kwargs,
+  )
+
+def vere_binary(copts = [], linkopts = [], **kwargs):
+  native.cc_binary(
+    copts = copts + select({
+        "//:debug": ["-O0", "-g3", "-DC3DBG"],
+        "//conditions:default": ["-O3"]
+    }) + select({
+        "//:lto": ['-flto'],
+        "//:thinlto": ['-flto=thin'],
+        "//conditions:default": []
+    }) + select({
+        "//:debug": [],
+        "@platforms//os:linux": ["-g"],
+        "//conditions:default": [],
+    }),
+    linkopts = linkopts + ['-g'] + select({
+        "//:lto": ['-flto'],
+        "//:thinlto": ['-flto=thin'],
+        "//conditions:default": []
+    }),
+    **kwargs,
+  )
--- a/bazel/toolchain/BUILD.bazel
+++ b/bazel/toolchain/BUILD.bazel
@ -50,7 +50,7 @@ _aarch64_gcc = "toolchain-gcc-linux-aarch64"

 cc_toolchain_config(
    name = "gcc-linux-aarch64-config",
-    ar = "{}/aarch64-linux-musl/bin/aarch64-linux-musl-ar".format(_install_prefix),
+    ar = "{}/aarch64-linux-musl/bin/aarch64-linux-musl-gcc-ar".format(_install_prefix),
    cc = "{}/aarch64-linux-musl/bin/aarch64-linux-musl-gcc".format(_install_prefix),
    cc_flags = [
        "-static",
@ -104,7 +104,7 @@ _x86_64_gcc = "toolchain-gcc-linux-x86_64"

 cc_toolchain_config(
    name = "gcc-linux-x86_64-config",
-    ar = "{}/x86_64-linux-musl/bin/x86_64-linux-musl-ar".format(_install_prefix),
+    ar = "{}/x86_64-linux-musl/bin/x86_64-linux-musl-gcc-ar".format(_install_prefix),
    cc = "{}/x86_64-linux-musl/bin/x86_64-linux-musl-gcc".format(_install_prefix),
    cc_flags = [
        "-static",
@ -279,7 +279,7 @@ cc_toolchain_config(
    # NOTE: building with `libtool` does not work on macOS due to lack of
    # support in the `configure_make` rule provided by `rules_foreign_cc`.
    # Therefore, we require setting `ar` as the archiver tool on macOS.
-    ar = "/usr/bin/ar",
+    ar = "/usr/local/opt/llvm@15/bin/llvm-ar",
    # By default, Bazel passes the `rcsD` flags to `ar`, but macOS's `ar`
    # implementation doesn't support `D`. We remove it with this attribute
    # and corresponding `ar_flags_feature` in `cfg.bzl`.
@ -288,7 +288,8 @@ cc_toolchain_config(
    cc = "/usr/local/opt/llvm@15/bin/clang",
    compiler = "clang",
    compiler_version = "//:clang_version",
-    ld = "/usr/bin/ld",
+    ld = "/usr/local/opt/llvm@15/bin/llvm-lld",
+    nm = "/usr/local/opt/llvm@15/bin/llvm-nm",
    sys_includes = [
        "/usr/local/Cellar/llvm@15/15.0.7/lib/clang/15.0.7/include",
        "/Library/Developer/CommandLineTools/SDKs/MacOSX12.sdk/usr/include",
--- a/pkg/c3/BUILD.bazel
+++ b/pkg/c3/BUILD.bazel
@ -2,7 +2,9 @@
 # LIBRARIES
 #

-cc_library(
+load("//bazel:common_settings.bzl", "vere_library")
+
+vere_library(
    name = "c3",
    srcs = glob(
        [
--- a/pkg/ent/BUILD.bazel
+++ b/pkg/ent/BUILD.bazel
@ -2,7 +2,9 @@
 # LIBRARIES
 #

-cc_library(
+load("//bazel:common_settings.bzl", "vere_library")
+
+vere_library(
    name = "ent",
    srcs = ["ent.c"],
    hdrs = ["ent.h"],
--- a/pkg/noun/BUILD.bazel
+++ b/pkg/noun/BUILD.bazel
@ -2,7 +2,9 @@
 # LIBRARIES
 #

-cc_library(
+load("//bazel:common_settings.bzl", "vere_library")
+
+vere_library(
    name = "noun",
    srcs = glob(
        [
--- a/pkg/ur/BUILD.bazel
+++ b/pkg/ur/BUILD.bazel
@ -2,7 +2,9 @@
 # LIBRARIES
 #

-cc_library(
+load("//bazel:common_settings.bzl", "vere_library")
+
+vere_library(
    name = "ur",
    srcs = [
        "bitstream.c",
--- a/pkg/urcrypt/BUILD.bazel
+++ b/pkg/urcrypt/BUILD.bazel
@ -2,7 +2,9 @@
 # LIBRARIES
 #

-cc_library(
+load("//bazel:common_settings.bzl", "vere_library")
+
+vere_library(
    name = "urcrypt",
    srcs = glob(
        [
--- a/pkg/urcrypt/ge-additions/BUILD.bazel
+++ b/pkg/urcrypt/ge-additions/BUILD.bazel
@ -1,4 +1,6 @@
-cc_library(
+load("//bazel:common_settings.bzl", "vere_library")
+
+vere_library(
    name = "ge-additions",
    srcs = ["ge-additions.c"],
    hdrs = ["ge-additions.h"],
--- a/pkg/vere/BUILD.bazel
+++ b/pkg/vere/BUILD.bazel
@ -2,6 +2,8 @@
 # GENERATED FILES
 #

+load("//bazel:common_settings.bzl", "vere_library", "vere_binary")
+
 # An approximation of `xxd -i` that runs on all platforms where Bash is
 # present. Generates a `.h` file that declares the array and array length as
 # `extern` global variables and a `.c` file containing the array and array
@ -86,7 +88,7 @@ genrule(
 # LIBRARIES
 #

-cc_library(
+vere_library(
    name = "vere",
    srcs = glob(
        [
@ -138,7 +140,7 @@ cc_library(
 # BINARIES
 #

-cc_binary(
+vere_binary(
    name = "urbit",
    srcs = [
        "main.c",