about summary refs log tree commit diff
path: root/absl/copts
diff options
context:
space:
mode:
authorAbseil Team <absl-team@google.com>2019-06-21T20·11-0700
committerGennadiy Rozental <rogeeff@google.com>2019-06-21T20·18-0400
commite9324d926a9189e222741fce6e676f0944661a72 (patch)
treea08568a709940c376454da34c9d8aac021378e5f /absl/copts
parent43ef2148c0936ebf7cb4be6b19927a9d9d145b8f (diff)
Export of internal Abseil changes.
--
7a6ff16a85beb730c172d5d25cf1b5e1be885c56 by Laramie Leavitt <lar@google.com>:

Internal change.

PiperOrigin-RevId: 254454546

--
ff8f9bafaefc26d451f576ea4a06d150aed63f6f by Andy Soffer <asoffer@google.com>:

Internal changes

PiperOrigin-RevId: 254451562

--
deefc5b651b479ce36f0b4ef203e119c0c8936f2 by CJ Johnson <johnsoncj@google.com>:

Account for subtracting unsigned values from the size of InlinedVector

PiperOrigin-RevId: 254450625

--
3c677316a27bcadc17e41957c809ca472d5fef14 by Andy Soffer <asoffer@google.com>:

Add C++17's std::make_from_tuple to absl/utility/utility.h

PiperOrigin-RevId: 254411573

--
4ee3536a918830eeec402a28fc31a62c7c90b940 by CJ Johnson <johnsoncj@google.com>:

Adds benchmark for the rest of the InlinedVector public API

PiperOrigin-RevId: 254408378

--
e5a21a00700ee83498ff1efbf649169756463ee4 by CJ Johnson <johnsoncj@google.com>:

Updates the definition of InlinedVector::shrink_to_fit() to be exception safe and adds exception safety tests for it.

PiperOrigin-RevId: 254401387

--
2ea82e72b86d82d78b4e4712a63a55981b53c64b by Laramie Leavitt <lar@google.com>:

Use absl::InsecureBitGen in place of std::mt19937
in tests absl/random/...distribution_test.cc

PiperOrigin-RevId: 254289444

--
fa099e02c413a7ffda732415e8105cad26a90337 by Andy Soffer <asoffer@google.com>:

Internal changes

PiperOrigin-RevId: 254286334

--
ce34b7f36933b30cfa35b9c9a5697a792b5666e4 by Andy Soffer <asoffer@google.com>:

Internal changes

PiperOrigin-RevId: 254273059

--
6f9c473da7c2090c2e85a37c5f00622e8a912a89 by Jorg Brown <jorg@google.com>:

Change absl::container_internal::CompressedTuple to instantiate its
internal Storage class with the name of the type it's holding, rather
than the name of the Tuple.  This is not an externally-visible change,
other than less compiler memory is used and less debug information is
generated.

PiperOrigin-RevId: 254269285

--
8bd3c186bf2fc0c55d8a2dd6f28a5327502c9fba by Andy Soffer <asoffer@google.com>:

Adding short-hand IntervalClosed for IntervalClosedClosed and IntervalOpen for
IntervalOpenOpen.

PiperOrigin-RevId: 254252419

--
ea957f99b6a04fccd42aa05605605f3b44b1ecfd by Abseil Team <absl-team@google.com>:

Do not directly use __SIZEOF_INT128__.

In order to avoid linker errors when building with clang-cl (__fixunsdfti, __udivti3 and __fixunssfti are undefined), this CL uses ABSL_HAVE_INTRINSIC_INT128 which is not defined for clang-cl.

PiperOrigin-RevId: 254250739

--
89ab385cd26b34d64130bce856253aaba96d2345 by Andy Soffer <asoffer@google.com>:

Internal changes

PiperOrigin-RevId: 254242321

--
cffc793d93eca6d6bdf7de733847b6ab4a255ae9 by CJ Johnson <johnsoncj@google.com>:

Adds benchmark for InlinedVector::reserve(size_type)

PiperOrigin-RevId: 254199226

--
c90c7a9fa3c8f0c9d5114036979548b055ea2f2a by Gennadiy Rozental <rogeeff@google.com>:

Import of CCTZ from GitHub.

PiperOrigin-RevId: 254072387

--
c4c388beae016c9570ab54ffa1d52660e4a85b7b by Laramie Leavitt <lar@google.com>:

Internal cleanup.

PiperOrigin-RevId: 254062381

--
d3c992e221cc74e5372d0c8fa410170b6a43c062 by Tom Manshreck <shreck@google.com>:

Update distributions.h to Abseil standards

PiperOrigin-RevId: 254054946

--
d15ad0035c34ef11b14fadc5a4a2d3ec415f5518 by CJ Johnson <johnsoncj@google.com>:

Removes functions with only one caller from the implementation details of InlinedVector by manually inlining the definitions

PiperOrigin-RevId: 254005427

--
2f37e807efc3a8ef1f4b539bdd379917d4151520 by Andy Soffer <asoffer@google.com>:

Initial release of Abseil Random

PiperOrigin-RevId: 253999861

--
24ed1694b6430791d781ed533a8f8ccf6cac5856 by CJ Johnson <johnsoncj@google.com>:

Updates the definition of InlinedVector::assign(...)/InlinedVector::operator=(...) to new, exception-safe implementations with exception safety tests to boot

PiperOrigin-RevId: 253993691

--
5613d95f5a7e34a535cfaeadce801441e990843e by CJ Johnson <johnsoncj@google.com>:

Adds benchmarks for InlinedVector::shrink_to_fit()

PiperOrigin-RevId: 253989647

--
2a96ddfdac40bbb8cb6a7f1aeab90917067c6e63 by Abseil Team <absl-team@google.com>:

Initial release of Abseil Random

PiperOrigin-RevId: 253927497

--
bf1aff8fc9ffa921ad74643e9525ecf25b0d8dc1 by Andy Soffer <asoffer@google.com>:

Initial release of Abseil Random

PiperOrigin-RevId: 253920512

--
bfc03f4a3dcda3cf3a4b84bdb84cda24e3394f41 by Laramie Leavitt <lar@google.com>:

Internal change.

PiperOrigin-RevId: 253886486

--
05036cfcc078ca7c5f581a00dfb0daed568cbb69 by Eric Fiselier <ericwf@google.com>:

Don't include `winsock2.h` because it drags in `windows.h` and friends,
and they define awful macros like OPAQUE, ERROR, and more. This has the
potential to break abseil users.

Instead we only forward declare `timeval` and require Windows users
include `winsock2.h` themselves. This is both inconsistent and poor QoI, but so
including 'windows.h' is bad too.

PiperOrigin-RevId: 253852615
GitOrigin-RevId: 7a6ff16a85beb730c172d5d25cf1b5e1be885c56
Change-Id: Icd6aff87da26f29ec8915da856f051129987cef6
Diffstat (limited to 'absl/copts')
-rw-r--r--absl/copts/GENERATED_AbseilCopts.cmake18
-rw-r--r--absl/copts/GENERATED_copts.bzl18
-rw-r--r--absl/copts/configure_copts.bzl41
-rw-r--r--absl/copts/copts.py14
4 files changed, 91 insertions, 0 deletions
diff --git a/absl/copts/GENERATED_AbseilCopts.cmake b/absl/copts/GENERATED_AbseilCopts.cmake
index c4948d42d495..88400e98b505 100644
--- a/absl/copts/GENERATED_AbseilCopts.cmake
+++ b/absl/copts/GENERATED_AbseilCopts.cmake
@@ -218,3 +218,21 @@ list(APPEND ABSL_MSVC_TEST_FLAGS
     "/wd4996"
     "/DNOMINMAX"
 )
+
+list(APPEND ABSL_RANDOM_HWAES_ARM32_FLAGS
+    "-mfpu=neon"
+)
+
+list(APPEND ABSL_RANDOM_HWAES_ARM64_FLAGS
+    "-march=armv8-a+crypto"
+)
+
+list(APPEND ABSL_RANDOM_HWAES_MSVC_X64_FLAGS
+    "/O2"
+    "/Ob2"
+)
+
+list(APPEND ABSL_RANDOM_HWAES_X64_FLAGS
+    "-maes"
+    "-msse4.1"
+)
diff --git a/absl/copts/GENERATED_copts.bzl b/absl/copts/GENERATED_copts.bzl
index 422b3a9b1c3a..d7edc93673da 100644
--- a/absl/copts/GENERATED_copts.bzl
+++ b/absl/copts/GENERATED_copts.bzl
@@ -219,3 +219,21 @@ ABSL_MSVC_TEST_FLAGS = [
     "/wd4996",
     "/DNOMINMAX",
 ]
+
+ABSL_RANDOM_HWAES_ARM32_FLAGS = [
+    "-mfpu=neon",
+]
+
+ABSL_RANDOM_HWAES_ARM64_FLAGS = [
+    "-march=armv8-a+crypto",
+]
+
+ABSL_RANDOM_HWAES_MSVC_X64_FLAGS = [
+    "/O2",
+    "/Ob2",
+]
+
+ABSL_RANDOM_HWAES_X64_FLAGS = [
+    "-maes",
+    "-msse4.1",
+]
diff --git a/absl/copts/configure_copts.bzl b/absl/copts/configure_copts.bzl
index 0015931762a4..8c4efe77c21a 100644
--- a/absl/copts/configure_copts.bzl
+++ b/absl/copts/configure_copts.bzl
@@ -16,6 +16,10 @@ load(
     "ABSL_MSVC_FLAGS",
     "ABSL_MSVC_LINKOPTS",
     "ABSL_MSVC_TEST_FLAGS",
+    "ABSL_RANDOM_HWAES_ARM32_FLAGS",
+    "ABSL_RANDOM_HWAES_ARM64_FLAGS",
+    "ABSL_RANDOM_HWAES_MSVC_X64_FLAGS",
+    "ABSL_RANDOM_HWAES_X64_FLAGS",
 )
 
 ABSL_DEFAULT_COPTS = select({
@@ -46,3 +50,40 @@ ABSL_DEFAULT_LINKOPTS = select({
     "//absl:windows": ABSL_MSVC_LINKOPTS,
     "//conditions:default": [],
 })
+
+# ABSL_RANDOM_RANDEN_COPTS blaze copts flags which are required by each
+# environment to build an accelerated RandenHwAes library.
+ABSL_RANDOM_RANDEN_COPTS = select({
+    # APPLE
+    ":cpu_darwin_x86_64": ABSL_RANDOM_HWAES_X64_FLAGS,
+    ":cpu_darwin": ABSL_RANDOM_HWAES_X64_FLAGS,
+    ":cpu_x64_windows_msvc": ABSL_RANDOM_HWAES_MSVC_X64_FLAGS,
+    ":cpu_x64_windows": ABSL_RANDOM_HWAES_MSVC_X64_FLAGS,
+    ":cpu_haswell": ABSL_RANDOM_HWAES_X64_FLAGS,
+    ":cpu_ppc": ["-mcrypto"],
+
+    # Supported by default or unsupported.
+    "//conditions:default": [],
+})
+
+# absl_random_randen_copts_init:
+#  Initialize the config targets based on cpu, os, etc. used to select
+#  the required values for ABSL_RANDOM_RANDEN_COPTS
+def absl_random_randen_copts_init():
+    """Initialize the config_settings used by ABSL_RANDOM_RANDEN_COPTS."""
+
+    # CPU configs.
+    # These configs have consistent flags to enable HWAES intsructions.
+    cpu_configs = [
+        "ppc",
+        "haswell",
+        "darwin_x86_64",
+        "darwin",
+        "x64_windows_msvc",
+        "x64_windows",
+    ]
+    for cpu in cpu_configs:
+        native.config_setting(
+            name = "cpu_%s" % cpu,
+            values = {"cpu": cpu},
+        )
diff --git a/absl/copts/copts.py b/absl/copts/copts.py
index 5bede34cb8e7..d850bb4f0553 100644
--- a/absl/copts/copts.py
+++ b/absl/copts/copts.py
@@ -199,4 +199,18 @@ COPT_VARS = {
         # Object file doesn't export any previously undefined symbols
         "-ignore:4221",
     ],
+    # "HWAES" is an abbreviation for "hardware AES" (AES - Advanced Encryption
+    # Standard). These flags are used for detecting whether or not the target
+    # architecture has hardware support for AES instructions which can be used
+    # to improve performance of some random bit generators.
+    "ABSL_RANDOM_HWAES_ARM64_FLAGS": ["-march=armv8-a+crypto"],
+    "ABSL_RANDOM_HWAES_ARM32_FLAGS": ["-mfpu=neon"],
+    "ABSL_RANDOM_HWAES_X64_FLAGS": [
+        "-maes",
+        "-msse4.1",
+    ],
+    "ABSL_RANDOM_HWAES_MSVC_X64_FLAGS": [
+        "/O2",  # Maximize speed
+        "/Ob2",  # Aggressive inlining
+    ],
 }