armsx/ARMSX3
0
0
Fork 0
mirror of https://github.com/ARMSX2/ARMSX3.git synced 2026-08-24 16:58:52 -07:00
Code Issues Packages Projects Releases Wiki Activity

Build: four release variants, and what the new one needed

Browse Source 
Splits the Android release into legacy / a11 / a13 / a15 so a device can take a
build matched to its CPU and OS instead of one binary suiting everything.
android/build-variants.sh drives all four from a single table of
(ndk, api, -march, apk suffix), and ConfigureCompiler.cmake takes -march per
variant rather than hardcoding one.

The legacy variant had never been compiled before: every release up to 0.7.2
was built at the gradle default of minSdk 33, so nothing had ever targeted a
lower API. Doing so turned up std::aligned_alloc, which is API 28+ -- below
that <cstdlib> does not declare it at all and the using-declaration fails to
resolve. posix_memalign is the older spelling and its result frees with plain
free(), so the rest of the header is unaffected. Kept even though legacy now
targets API 30, because it costs nothing and the next person to try a lower
floor should not rediscover it.

legacy targets armv8.1-a, which is the floor this codebase compiles at rather
than a preference: util/simd.hpp uses SQRDMLAH (v8.1 RDMA) and util/asm.hpp
has inline LSE atomics, so armv8-a does not build. Its value is cores that are
ARMv8.2 without the OPTIONAL fp16 and dotprod extensions the other three
variants require. Cortex-A53/A72/A73 class parts stay out of reach until those
two paths gain fallbacks.
This commit is contained in:
jpolo1224
2026-08-15 01:23:53 -04:00
parent 0a9fd15b57
commit eb54f9b75a
4 changed files with 231 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
buildfiles/cmake/ConfigureCompiler.cmake
+21 -2
View File
@@ -22,7 +22,18 @@ else()
if (APPLE)
check_cxx_compiler_flag("-march=armv8.4-a" COMPILER_ARM)
else()
check_cxx_compiler_flag("-march=armv8.1-a" COMPILER_ARM)
# Per-variant, set by android/build-variants.sh. Default is the standard/new target;
# the legacy variant passes armv8-a so it still runs on Cortex-A53/A73-class cores.
# (clang spells the 8.0 baseline "armv8-a" and rejects "armv8.0-a" outright.)
#
# armv8.0, NOT armv8.1: ARMv8.1 mandates LSE atomics and the mobile core line skipped that
# revision entirely -- A53/A72/A73 are 8.0 and A55/A75 onwards are 8.2. An armv8.1 build
# therefore faults on exactly the old devices a fallback exists for, while running only on
# parts that already handle 8.2. It is 8.0 or it is pointless.
if (NOT DEFINED ARMSX3_ARM_MARCH)
set(ARMSX3_ARM_MARCH "armv8.2-a+dotprod+fp16")
endif()
check_cxx_compiler_flag("-march=${ARMSX3_ARM_MARCH}" COMPILER_ARM)
endif()
add_compile_options(-Wall)
@@ -37,7 +48,15 @@ else()
if (APPLE)
add_compile_options(-march=armv8.4-a)
else()
add_compile_options(-march=armv8.1-a)
# Default armv8.2-a+dotprod+fp16. dotprod is only mandatory from 8.4 and FP16
# arithmetic is optional in 8.2, so "armv8.2-a" alone guarantees neither and both
# are named. LSE atomics come from the 8.1 baseline and are already implied.
#
# This is a hard floor, not a preference: the instructions can appear anywhere in
# the C++, and a device without them faults rather than falling back. Cortex-A55/A75
# and later, i.e. Snapdragon 845 (2018) onward. The legacy variant overrides this
# with armv8.1-a to stay below that line.
add_compile_options(-march=${ARMSX3_ARM_MARCH})
endif()
elseif(COMPILER_X86)
# Some compilers will set both X86 and ARM, so check explicitly for ARM first