|Age||Commit message (Collapse)||Author||Files||Lines|
The accelerated convolutional decoder uses SSSE3 instructions such
as PSIGNW (via _mm_sign_epi16) which go beyond what SSE3 offers. So
let's make sure we use the right compiler flag (-mssse3) and also the
right runtime check.
Without this patch, we would use illegal instructions e.g. on Opteron
Gen3 such as Opteron 2427, which are also used as build.opensuse.org
build hosts (build31 through build36) where we wouldn't pass "make
check" as a result.
HAVE_AVX2, HAVE_SSE3 and HAVE_SSE4_1 were never defined if CPU
architecture doesn't match the (86*|x86_64*|amd64*) condition.
This change adds a check whether compiler supports some SIMD
(Single Instruction, Multiple Data) instructions. It is enabled
by default, and can be disabled by configure option --disable-simd.
The check macro is based on the AX_EXT from autoconf-archive:
And depends on the ax_check_compile_flag macro:
Currently only the following SIMD extensions are being checked:
AVX2, SSE3, SSE4.1, but adding others is also possible. All found
extensions are being defined in the 'config.h' header.
Even when removing AC_CONFIG_MACRO_DIR aclocal insisted that it
needs to have a m4 directory. Make it happy by providing one. As git
is not tracking directories I needed to add a dummy file.