incompatible types when initializing type ‘float32x4_t’ using type ‘float32x2_t’

通过修改arm_neon.h文件中的vmlaq_lane_f32宏定义来解决问题。

在aarch64平台里用gcc编译libopus时，出现错误，错误表现如下：

In file included from celt/arm/celt_neon_intr.c:37:0:
celt/arm/celt_neon_intr.c: In function ‘xcorr_kernel_neon_float’:
celt/arm/celt_neon_intr.c:137:14: error: incompatible types when initializing type ‘float32x4_t’ using type ‘float32x2_t’
       SUMM = vmlaq_lane_f32(SUMM, YY[0], vget_low_f32(XX[0]), 0);
              ^
celt/arm/celt_neon_intr.c:139:14: error: incompatible types when initializing type ‘float32x4_t’ using type ‘float32x2_t’
       SUMM = vmlaq_lane_f32(SUMM, YEXT[0], vget_low_f32(XX[0]), 1);
              ^
celt/arm/celt_neon_intr.c:141:14: error: incompatible types when initializing type ‘float32x4_t’ using type ‘float32x2_t’
       SUMM = vmlaq_lane_f32(SUMM, YEXT[1], vget_high_f32(XX[0]), 0);
              ^
celt/arm/celt_neon_intr.c:143:14: error: incompatible types when initializing type ‘float32x4_t’ using type ‘float32x2_t’
       SUMM = vmlaq_lane_f32(SUMM, YEXT[2], vget_high_f32(XX[0]), 1);
              ^
celt/arm/celt_neon_intr.c:145:14: error: incompatible types when initializing type ‘float32x4_t’ using type ‘float32x2_t’
       SUMM = vmlaq_lane_f32(SUMM, YY[1], vget_low_f32(XX[1]), 0);
              ^
celt/arm/celt_neon_intr.c:147:14: error: incompatible types when initializing type ‘float32x4_t’ using type ‘float32x2_t’
       SUMM = vmlaq_lane_f32(SUMM, YEXT[0], vget_low_f32(XX[1]), 1);
              ^
celt/arm/celt_neon_intr.c:149:14: error: incompatible types when initializing type ‘float32x4_t’ using type ‘float32x2_t’
       SUMM = vmlaq_lane_f32(SUMM, YEXT[1], vget_high_f32(XX[1]), 0);
              ^
celt/arm/celt_neon_intr.c:151:14: error: incompatible types when initializing type ‘float32x4_t’ using type ‘float32x2_t’
       SUMM = vmlaq_lane_f32(SUMM, YEXT[2], vget_high_f32(XX[1]), 1);
              ^
celt/arm/celt_neon_intr.c:170:14: error: incompatible types when initializing type ‘float32x4_t’ using type ‘float32x2_t’
       SUMM = vmlaq_lane_f32(SUMM, YY[0], vget_low_f32(XX[0]), 0);
              ^
celt/arm/celt_neon_intr.c:172:14: error: incompatible types when initializing type ‘float32x4_t’ using type ‘float32x2_t’
       SUMM = vmlaq_lane_f32(SUMM, YEXT[0], vget_low_f32(XX[0]), 1);
              ^
celt/arm/celt_neon_intr.c:174:14: error: incompatible types when initializing type ‘float32x4_t’ using type ‘float32x2_t’
       SUMM = vmlaq_lane_f32(SUMM, YEXT[1], vget_high_f32(XX[0]), 0);
              ^
celt/arm/celt_neon_intr.c:176:14: error: incompatible types when initializing type ‘float32x4_t’ using type ‘float32x2_t’
       SUMM = vmlaq_lane_f32(SUMM, YEXT[2], vget_high_f32(XX[0]), 1);
              ^
celt/arm/celt_neon_intr.c:184:14: error: incompatible types when initializing type ‘float32x4_t’ using type ‘float32x2_t’
       SUMM = vmlaq_lane_f32(SUMM, YY[0], XX_2, 0);
              ^
celt/arm/celt_neon_intr.c:189:11: error: incompatible types when initializing type ‘float32x4_t’ using type ‘float32x2_t’
    SUMM = vmlaq_lane_f32(SUMM, YY[0], XX_2, 0);
           ^
make[2]: *** [celt/arm/celt_neon_intr.lo] Error 1

反复出现如下错误：

celt/arm/celt_neon_intr.c: In function ‘xcorr_kernel_neon_float’:
celt/arm/celt_neon_intr.c:137:14: error: incompatible types when initializing type ‘float32x4_t’ using type ‘float32x2_t’
       SUMM = vmlaq_lane_f32(SUMM, YY[0], vget_low_f32(XX[0]), 0);

      这个是GCC的系统的错误，是因为 arm_neon.h 头里对vmlaq_lane_f32  这个宏的错误定义所致，当前系统的GCC版本是：

 gcc version 4.8.5 20150623 (Red Hat 4.8.5-44) (GCC),
该版本所带的头文件 /usr/lib/gcc/aarch64-redhat-linux/4.8.2/include/arm_neon.h 里定义 vmlaq_lane_f32 有错误，因此可以通过修改该文件的vmlaq_lane_f32定义来解决问题，反正/usr/lib/gcc/aarch64-redhat-linux/4.8.2/include/arm_neon.h 本身对vmlaq_lane_f32 的定义有错误，所以修正此处也不会给系统带来影响，笔者修正后vmlaq_lane_f32 的定义如下：
 

#define vmlaq_lane_f32(a, b, c, d)                                      \
  __extension__                                                         \
    ({                                                                  \
       float32x2_t c_ = (c);                                            \
       float32x4_t b_ = (b);                                            \
       float32x4_t a_ = (a);                                            \
       float32x4_t result;                                              \
       float32x4_t t1;                                                  \
       __asm__ ("fmul %1.4s, %3.4s, %4.s[%5]; fadd %0.4s, %0.4s, %1.4s" \
                : "=w"(result), "=w"(t1)                                \
                : "0"(a_), "w"(b_), "w"(c_), "i"(d)                     \
                : /* No clobbers */);                                   \
       result;                                                          \
     })

原始行的代码 float32x4_t c_ = (c); 被修正为 float32x2_t c_ = (c);

因为 vmlaq_lane_f32 在 GCC 4.8.5 里只是头文件里定义的宏，不是编译后的库文件中符号，因此可以放心大胆的修改该头文件来解决这个问题

还有一处错误，对 vmlaq_lane_s32 的定义也是有问题的，修改如下：

#define vmlaq_lane_s32(a, b, c, d)                                      \
  __extension__                                                         \
    ({                                                                  \
       int32x2_t c_ = (c);                                              \
       int32x4_t b_ = (b);                                              \
       int32x4_t a_ = (a);                                              \
       int32x4_t result;                                                \
       __asm__ ("mla %0.4s, %2.4s, %3.s[%4]"                            \
                : "=w"(result)                                          \
                : "0"(a_), "w"(b_), "w"(c_), "i"(d)                     \
                : /* No clobbers */);                                   \
       result;                                                          \
     })

      当然如果你比较悠闲，有足够的时间或者对修改系统头文件有担惊受怕的心理，你可以通过编译升级GCC到7.3来解决编译问题，GCC 7.3 对 已经将 vmlaq_lane_f32 改成内联函数了。