Bug in openEuler/gcc (10.3.1) support for ARM SVE under O0 compiler level

问题描述：gcc编译器在O0下编译sve代码时，生成的汇编不符合预期
复现步骤 demo.c
#include <arm_sve.h>

static inline svuint64_t test(svbool_t pg, svuint64_t x, svbool_t cmp)
{
    svuint64_t res = svsel_u64(cmp, x, svadd_n_u64_x(pg, x, 10ULL));
    res = svsub_u64_x(pg, res, x);
    return res;
}

int main(int argc, char *argv[])
{
    uint64_t xInit[] = {
        0, 1, 2, 3,
    };
    svbool_t pg = svptrue_b64();
    svuint64_t x = svld1_u64(pg, xInit);
    svbool_t cmp = svcmpge_n_u64(pg, x, 2ULL);
    svuint64_t  res = test(pg, x, cmp);
    return 0;
}

编译命令：
gcc -std=c99 -O0 -march=armv8.3-a+fp+sve -o demo.exe demo.c

gcc 版本信息：
Using built-in specs.
COLLECT_GCC=gcc
COLLECT_LTO_WRAPPER=/usr/libexec/gcc/aarch64-linux-gnu/10.3.1/lto-wrapper
Target: aarch64-linux-gnu
Configured with: ../configure --prefix=/usr --mandir=/usr/share/man --infodir=/usr/share/info --enable-shared --enable-threads=posix --enable-checking=release --with-system-zlib --enable-__cxa_atexit --disable-libunwind-exceptions --enable-gnu-unique-object --enable-linker-build-id --with-linker-hash-style=gnu --enable-languages=c,c++,objc,obj-c++,fortran,lto --enable-plugin --enable-initfini-array --disable-libgcj --without-isl --without-cloog --enable-gnu-indirect-function --build=aarch64-linux-gnu --with-stage1-ldflags=' -Wl,-z,relro,-z,now' --with-boot-ldflags=' -Wl,-z,relro,-z,now' --disable-bootstrap --with-multilib-list=lp64 --enable-bolt
Thread model: posix
Supported LTO compression algorithms: zlib
gcc version 10.3.1 (GCC)

问题现象：
生成的exe反汇编后sel指令被优化，处理逻辑错误。
0000000000400854 <test>:
  400854:	043f57bf 	addvl	sp, sp, #-3
  400858:	e5811fe0 	str	p0, [sp, #15, mul vl]
  40085c:	2518e3e0 	ptrue	p0.b
  400860:	e5e0e3e0 	st1d	{z0.d}, p0, [sp]
  400864:	e5811be1 	str	p1, [sp, #14, mul vl]
  400868:	2518e3e0 	ptrue	p0.b
  40086c:	a5e2a3e1 	ld1d	{z1.d}, p0/z, [sp, #2, mul vl]
  400870:	2518e3e0 	ptrue	p0.b
  400874:	a5e0a3e0 	ld1d	{z0.d}, p0/z, [sp]
  400878:	04e00420 	sub	z0.d, z1.d, z0.d
  40087c:	2518e3e0 	ptrue	p0.b
  400880:	e5e2e3e0 	st1d	{z0.d}, p0, [sp, #2, mul vl]
  400884:	2518e3e0 	ptrue	p0.b
  400888:	a5e2a3e0 	ld1d	{z0.d}, p0/z, [sp, #2, mul vl]
  40088c:	043f507f 	addvl	sp, sp, #3
  400890:	d65f03c0 	ret

src-openEuler/gcc
Closed

Content Risk Flag

Comments (6)

src-openEuler/gccClosed .gitee-modal { width: 500px !important; }

Content Risk Flag