[compiler-rt] Fix build errors when using gcc on LoongArch

[tangyouling]

Summary

• GCC does not recognize $fcsr0, uses $r0 instead.
• GCC does not implement __builtin_thread_pointer, which can be obtained directly through $tp.

[SixWeining]

LGTM

compiler-rt/lib/builtins/loongarch/fp_mode.c

 #define LOONGARCH_RMODE_MASK (LOONGARCH_TONEAREST | LOONGARCH_TOWARDZERO | \
                               LOONGARCH_UPWARD | LOONGARCH_DOWNWARD)
 
 #define LOONGARCH_INEXACT    0x10000
 
 CRT_FE_ROUND_MODE __fe_getround(void) {
 #if __loongarch_frlen != 0
   int fcsr;
+#  ifdef __clang__
   __asm__ __volatile__("movfcsr2gr %0, $fcsr0" : "=r" (fcsr));
+#  else
+  __asm__ __volatile__("movfcsr2gr %0, $r0" : "=r" (fcsr));
+#  endif
   fcsr &= LOONGARCH_RMODE_MASK;
   switch (fcsr) {
   case LOONGARCH_TOWARDZERO:
     return CRT_FE_TOWARDZERO;
   case LOONGARCH_DOWNWARD:
     return CRT_FE_DOWNWARD;
   case LOONGARCH_UPWARD:
     return CRT_FE_UPWARD;
   case LOONGARCH_TONEAREST:
   default:
     return CRT_FE_TONEAREST;
   }
 #else
   return CRT_FE_TONEAREST;
 #endif
 }
 
 int __fe_raise_inexact(void) {
 #if __loongarch_frlen != 0
   int fcsr;
+#  ifdef __clang__
   __asm__ __volatile__("movfcsr2gr %0, $fcsr0" : "=r" (fcsr));
   __asm__ __volatile__(
       "movgr2fcsr $fcsr0, %0" :: "r" (fcsr | LOONGARCH_INEXACT));
+#  else
+  __asm__ __volatile__("movfcsr2gr %0, $r0" : "=r" (fcsr));
+  __asm__ __volatile__(
+      "movgr2fcsr $r0, %0" :: "r" (fcsr | LOONGARCH_INEXACT));
+#  endif
 #endif
   return 0;
 }

compiler-rt/lib/sanitizer_common/sanitizer_linux_libcdep.cpp

#  endif
   *size = g_tls_size;
   *addr -= *size;
   *addr += ThreadDescriptorSize();
 #elif SANITIZER_GLIBC && defined(__aarch64__)
   *addr = reinterpret_cast<uptr>(__builtin_thread_pointer()) -
           ThreadDescriptorSize();
   *size = g_tls_size + ThreadDescriptorSize();
 #elif SANITIZER_GLIBC && defined(__loongarch__)
+#  ifdef __clang__
   *addr = reinterpret_cast<uptr>(__builtin_thread_pointer()) -
           ThreadDescriptorSize();
+#  else
+  asm("or %0,$tp,$zero" : "=r"(*addr));
+  *addr -= ThreadDescriptorSize();
+#  endif
   *size = g_tls_size + ThreadDescriptorSize();
 #elif SANITIZER_GLIBC && defined(__powerpc64__)
   // Workaround for glibc<2.25(?). 2.27 is known to not need this.
   uptr tp;
   asm("addi %0,13,-0x7000" : "=r"(tp));
   const uptr pre_tcb_size = TlsPreTcbSize();
   *addr = tp - pre_tcb_size;
   *size = g_tls_size + pre_tcb_size;