security/hashcat: fix build on powerpc64le
Build with MAINTAINER_MODE=1 so the bridges don't get -march=native.
Clang 21 rejects it on powerpc and packages shouldn't use it anyway.
lang/pocl: backport fix for crashes with big work-groups
Worker threads were created with the default 2 MB stack, which is too
small for kernels that keep a lot of private data per work-item.
Programs like hashcat crash because of that.
Obtained from: https://github.com/pocl/pocl/commit/0222df2be566a3e74816f6ba9856efae7d1a8d80
editors/gram: fix build on powerpc64le
Pull in the upstream fix for KINFO_FILE_SIZE, which is only defined on
amd64 and i386. The vendored minidump-writer and crash-handler crates
also lack powerpc64 in their arch switches.
Approved by: portmgr (blanket)
devel/simde: fix the roundeven fallback on powerpc64
clang turns __builtin_roundeven into a roundeven() call when there is no
native rounding instruction and our libm doesn't have it, so consumers
fail to link on powerpc64. Backport the glibc check from simde
6b772fe4e9a0 and cb5b957aa7db.
Approved by: portmgr (blanket)
audio/surge-synthesizer-xt-lv2: fix build on powerpc64
The bundled simde uses __builtin_roundevenf when there is no VSX and
clang turns it into a call to roundevenf, which our libm doesn't have.
Backport the glibc check from simde 6b772fe4e9a0 and cb5b957aa7db.
Approved by: portmgr (blanket)
sysutils/mise: fix build and runtime on powerpc64
sonic-number refuses to build on big-endian and sonic-simd's portable
fallback puts the bitmask bits in the wrong order there, so sonic-rs
runs off the buffer on the first string it escapes. Backport the fix
from cloudwego/sonic-rs#265.
www/waterfox: fix build on powerpc64
The WASM sandbox isn't available on big-endian, so only pass
--with-wasi-sysroot and pull in the wasi dependencies on other
architectures, like www/firefox does.
math/kalc: fix build on powerpc64
The bundled GMP fails its own make check with clang 21: the PowerPC inline
asm only clobbers "cc", so clang keeps a carry in XER across it and
mpn_div_qr_1 comes out wrong. Declare "xer" too.
security/agave: fix build on powerpc64
blstrs and merlin refuse to build on big-endian targets. blstrs'
Scalar::to_bytes_le() dumps the u64 limbs as native bytes, merlin feeds
its keccak state to f1600 in native byte order, and the blst crate
underneath uses 32-bit limbs in its C code on everything but x86_64 and
aarch64 while the Rust bindings declare them as u64, which only agrees
on little-endian. Use blst's own conversion in blstrs and backport
supranational/blst#109 and zkcrypto/merlin#5.
games/pioneer: fix build on powerpc64
Pull in the upstream fix for big-endian hosts.
The models only compile with assimp 6.0.6 or newer on big-endian, older
versions misread the COLLADA input offsets there.
Approved by: portmgr (blanket)
mail/oama: fix build on powerpc64
The simdutf bundled with text enables its POWER code whenever AltiVec is
available, but that code needs POWER8. Backport simdutf 573d42d7, which
also checks for POWER8.
misc/py-vllm: fix build on powerpc64
oneDNN builds its ppc64 kernels with -mcpu=native, which clang on FreeBSD
resolves to a plain ppc64 without VSX. Pass it the same -mcpu the rest of
the extension is built with. Bump PORTREVISION since this also changes
the oneDNN flags on powerpc64le.
biology/mmseqs2: fix build and alignment on powerpc64
The bundled SIMDe calls __builtin_roundevenf, which becomes a roundevenf()
libm call without VSX. FreeBSD doesn't have roundevenf(), so use SIMDe's
inline fallback instead, like SIMDe upstream does now.
simd_hmax8_sse() and simd_hmax16_sse() read the _mm_minpos_epu16 result
with _mm_cvtsi128_si32, which on big-endian gives the lane index instead
of the minimum, so every alignment died. Use _mm_extract_epi16 instead.
This is fixed upstream in 0e90db6dc459, which doesn't apply to 17-b804f.
emulators/einstein-newton: fix build on powerpc64
The big-endian code paths still pass a long-dropped size argument to
memcpy and leave the sound buffer member unused, which -Werror rejects.
lang/gcc17-devel: fix build on powerpc64
GCC 17 now checks the exit status of the stage comparison, and
bootstrap-lto-noplugin.mk uses /bin/true, which FreeBSD doesn't have.
Use the shell builtin instead. Also list libitm.info on all arches, it
gets built on powerpc too.
mail/stalwart: fix build and runtime on powerpc64
sonic-number refuses to build on big-endian and sonic-simd's portable
fallback puts the bitmask bits in the wrong order there, so sonic-rs
runs off the buffer on the first string it escapes. Backport the fix
from cloudwego/sonic-rs#265.
www/trafficserver: fix build on powerpc64
The big-endian branch of the bundled libswoc never compiled, it was dead
code until trafficserver switched the endianness check to __BYTE_ORDER__.
sysutils/passwordsafe: fix build on powerpc64
The big-endian code initializes a variable length array, which clang
rejects. Backport the upstream fix (8904886a592b).
mail/stalwart-migrate: fix build and runtime on powerpc64
sonic-number refuses to build on big-endian and sonic-simd's portable
fallback puts the bitmask bits in the wrong order there, so sonic-rs
runs off the buffer on the first string it escapes. Backport the fix
from cloudwego/sonic-rs#265.
emulators/qemu-user-static: fix runtime on powerpc64
machdep.cacheline_size is an int and the code read it into a long, so on
big-endian hosts every emulator died in init_cache_info() before doing
anything. Take the upstream fix.
graphics/crucible: fix build on powerpc64
util/string.h refuses big-endian hosts with an #error, but nothing
depends on the bit-field layout it guards.
games/ivan: fix build and startup crash on powerpc64
The big-endian branch in graphics.cpp was only ever written for SDL1, and
the MIDI parser byte-swaps file fields as if the host were little-endian.
emulators/qemu-user-static: enable on powerpc64le
The code already supports little-endian POWER hosts, only the port kept
it out. Point the little-endian build at the ppc64 host directory like
emulators/qemu does.
textproc/evtx: fix build and runtime on powerpc64
sonic-number refuses to build on big-endian and sonic-simd's portable
fallback puts the bitmask bits in the wrong order there, so sonic-rs
runs off the buffer on the first string it escapes. Backport the fix
from cloudwego/sonic-rs#265.
www/qt6-webengine: fix the build on powerpc64le
Qt 6.11.3 ships a V8 where frames.cc already uses
WasmLiftoffSetupFrameConstants::kCallingPCOffset (V8 dec5f6bdbe68) but the
PPC frame constants did not get it yet (V8 f285826766bd). Add the one line.
misc/codex: fix gn gen on powerpc64*
The partition_alloc revision rusty_v8 pins doesn't know ppc64 yet, so gn
gen dies with "Unknown CPU: ppc64" since V8 started importing
partition_alloc.gni. Backport the upstream fix (Chromium 8fd8e6027af0).
mksnapshot (and so the build) still fails most of the time under the
52-bit ASLR of 15.2 kernels: V8 keeps cppgc pointers in 48 bits. That
one is for the kernel (D61331).
japanese/fcitx5-hazkey: fix build on powerpc64le
The bundled llama.cpp still has register in the POWER9 fp16 helpers,
which clang rejects in C++17. Pull in the upstream fix.
audio/maolan: fix build on powerpc64
The big-endian code in the hi_sparse_bitset crate doesn't compile, it
was never built upstream. Implement it so the serialized format is the
same as on little-endian.