AMD或Intel上编译出来的程序，可以跑在海光上吗？

        在上一篇博文《海光处理器与AMD Zen1的指令差异-CSDN博客》中发现，海光相比AMD，缺失了一些指令集。

        那么在AMD或Intel上编译出来的程序，可以跑在海光上吗？

        这个问题的关键，在于编译器默认使用哪些指令来编译程序。以CentOS8.4为例，它自带的gcc版本为gcc version 8.4.1 20200928：

[root@CentOS8 ~]# gcc -v Using built-in specs. COLLECT_GCC=/usr/bin/gcc COLLECT_LTO_WRAPPER=/usr/libexec/gcc/x86_64-redhat-linux/8/lto-wrapper OFFLOAD_TARGET_NAMES=nvptx-none OFFLOAD_TARGET_DEFAULT=1 Target: x86_64-redhat-linux Configured with: ../configure --enable-bootstrap --enable-languages=c,c++,fortran,lto --prefix=/usr --mandir=/usr/share/man --infodir=/usr/share/info --with-bugurl=http://bugzilla.redhat.com/bugzilla --enable-shared --enable-threads=posix --enable-checking=release --enable-multilib --with-system-zlib --enable-__cxa_atexit --disable-libunwind-exceptions --enable-gnu-unique-object --enable-linker-build-id --with-gcc-major-version-only --with-linker-hash-style=gnu --enable-plugin --enable-initfini-array --with-isl --disable-libmpx --enable-offload-targets=nvptx-none --without-cuda-driver --enable-gnu-indirect-function --enable-cet --with-tune=generic --with-arch_32=x86-64 --build=x86_64-redhat-linux Thread model: posix gcc version 8.4.1 20200928 (Red Hat 8.4.1-1) (GCC)

        gcc在编译的时候，可通过-march=来指定要使用的指令范围，默认值为-march=x86-64，这个默认值是构建gcc自身时指定的，即上表中的--with-arch_32=x86-64。只要指定了--with-arch_32和--with-arch_64其中一个并且处理器支持64位，另外一个变量都会被指定为x86-64。

        接下来看一下-march=x86-64代表的指令范围。

[root@CentOS8 ~]# gcc -march=x86-64 -c -Q --help=target | grep -i enabled   -m128bit-long-double                  [enabled]   -m64                                  [enabled]   -m80387                               [enabled]   -malign-stringops                     [enabled]   -mavx256-split-unaligned-load         [enabled]   -mavx256-split-unaligned-store        [enabled]   -mfancy-math-387                      [enabled]   -mfp-ret-in-387                       [enabled]   -mfxsr                                [enabled]   -mglibc                               [enabled]   -mhard-float                          [enabled]   -mieee-fp                             [enabled]   -mlong-double-80                      [enabled]   -mmmx                                 [enabled]   -mno-sse4                             [enabled]   -mpush-args                           [enabled]   -mred-zone                            [enabled]   -msse                                 [enabled]   -msse2                                [enabled]   -mstv                                 [enabled]   -mtls-direct-seg-refs                 [enabled]   -mvzeroupper                          [enabled]

        这些指令代表什么意思呢？

-march=x86-64 of GCC8.4.1(CentOS8.4)	x86-64	Instructions
m128bit-long-double	[enabled]	sizeof(long double) is 16.
m64	[enabled]	Generate 64bit x86-64 code.
m80387	[enabled]	Use hardware fp.
malign-stringops	[enabled]	Align destination of the string operations.
march=	x86-64	Generate code for given CPU.
mavx256-split-unaligned-load	[enabled]	d Split 32-byte AVX unaligned load.
mavx256-split-unaligned-store	[enabled]	re Split 32-byte AVX unaligned store.
mfancy-math-387	[enabled]	Generate sin, cos, sqrt for FPU.
mfp-ret-in-387	[enabled]	Return values of functions in FPU registers.
mfxsr	[enabled]	Support FXSAVE and FXRSTOR instructions.
mglibc	[enabled]	Use GNU C library.
mhard-float	[enabled]	Use hardware fp.
mieee-fp	[enabled]	Use IEEE math for fp comparisons.
mlong-double-80	[enabled]	Use 80-bit long double.
mmmx	[enabled]	Support MMX built-in functions.
mno-sse4	[enabled]	Do not support SSE4.1 and SSE4.2 built-in functions and code generation.
mpush-args	[enabled]	Use push instructions to save outgoing arguments.
mred-zone	[enabled]	Use red-zone in the x86-64 code.
msse	[enabled]	Support MMX and SSE built-in functions and code generation.
msse2	[enabled]	Support MMX, SSE and SSE2 built-in functions and code generation.
mstv	[enabled]	Disable Scalar to Vector optimization pass transforming 64-bit integer computations into a vector ones.
mtls-direct-seg-refs	[enabled]	Use direct references against %gs when accessing tls data.
mvzeroupper	[enabled]	Generate vzeroupper instruction before a transfer of control flow out of the function.

        很明显，这个-march=x86-64代表的指令范围很小，海光全部都支持。所以，默认情况下，在AMD上编译的程序，依然可以跑在海光上。 

         如果gcc编译的时候指定-march=znver1，那么海光可以运行吗？答案是否定的，因为至少有2个指令集海光是不支持的：aes和pclmul。

        只有一种特例：那就是程序本身并不涉及加解密，编译之后的汇编代码不包括aes和pclmul相关指令，即使编译的时候指定-march=znver1，那么这个程序还是可以跑在海光上的。

-march=??? of GCC8.4.1(CentOS8.4)	x86-64	znver1
m128bit-long-double	[enabled]	[enabled]
m64	[enabled]	[enabled]
m80387	[enabled]	[enabled]
mabm	[disabled]	[enabled]
madx	[disabled]	[enabled]
maes	[disabled]	[enabled]
malign-stringops	[enabled]	[enabled]
march=	x86-64	znver1
mavx	[disabled]	[enabled]
mavx2	[disabled]	[enabled]
mavx256-split-unaligned-store	[enabled]	[enabled]
mbmi	[disabled]	[enabled]
mbmi2	[disabled]	[enabled]
mclflushopt	[disabled]	[enabled]
mclzero	[disabled]	[enabled]
mcx16	[disabled]	[enabled]
mf16c	[disabled]	[enabled]
mfancy-math-387	[enabled]	[enabled]
mfma	[disabled]	[enabled]
mfp-ret-in-387	[enabled]	[enabled]
mfsgsbase	[disabled]	[enabled]
mfxsr	[enabled]	[enabled]
mglibc	[enabled]	[enabled]
mhard-float	[enabled]	[enabled]
mieee-fp	[enabled]	[enabled]
mlong-double-80	[enabled]	[enabled]
mlzcnt	[disabled]	[enabled]
mmmx	[enabled]	[enabled]
mmovbe	[disabled]	[enabled]
mmwaitx	[disabled]	[enabled]
mpclmul	[disabled]	[enabled]
mpopcnt	[disabled]	[enabled]
mprfchw	[disabled]	[enabled]
mpush-args	[enabled]	[enabled]
mrdrnd	[disabled]	[enabled]
mrdseed	[disabled]	[enabled]
mred-zone	[enabled]	[enabled]
msahf	[disabled]	[enabled]
msha	[disabled]	[enabled]
msse	[enabled]	[enabled]
msse2	[enabled]	[enabled]
msse3	[disabled]	[enabled]
msse4	[disabled]	[enabled]
msse4.1	[disabled]	[enabled]
msse4.2	[disabled]	[enabled]
msse4a	[disabled]	[enabled]
mssse3	[disabled]	[enabled]
mstv	[enabled]	[enabled]
mtls-direct-seg-refs	[enabled]	[enabled]
mtune=	generic	znver1
mvzeroupper	[enabled]	[enabled]
mxsave	[disabled]	[enabled]
mxsavec	[disabled]	[enabled]
mxsaveopt	[disabled]	[enabled]
mxsaves	[disabled]	[enabled]

如果你觉得有用，请点赞收藏哈！