refine: improve RISC-V support with cpu_relax and const correctness

- Add RISC-V cpu_relax() implementation using fence.i instruction
- Fix const correctness in AtomicInteger128 mutex operations

Co-authored-by: gong-flying <gongxiaofei24@iscas.ac.cn>

Author: Dayuxiaoshui

src/bthread/processor.h

@@ -28,6 +28,8 @@
 # ifndef cpu_relax
 #if defined(ARCH_CPU_ARM_FAMILY)
 # define cpu_relax() asm volatile("yield\n": : :"memory")
+#elif defined(ARCH_CPU_RISCV_FAMILY)
+# define cpu_relax() asm volatile("fence.i\n": : :"memory")
 #elif defined(ARCH_CPU_LOONGARCH64_FAMILY)
 # define cpu_relax() asm volatile("nop\n": : :"memory");
 #else

src/bthread/task_group.cpp

@@ -101,7 +101,7 @@ AtomicInteger128::Value AtomicInteger128::load() const {
     return {value[0], value[1]};
 #else // __x86_64__ || __ARM_NEON
     // RISC-V and other architectures use mutex fallback
-    BAIDU_SCOPED_LOCK(_mutex);
+    BAIDU_SCOPED_LOCK(const_cast<FastPthreadMutex&>(_mutex));
     return _value;
 #endif // __x86_64__ || __ARM_NEON
 }
@@ -115,7 +115,7 @@ void AtomicInteger128::store(Value value) {
     vst1q_s64(reinterpret_cast<int64_t*>(&_value), v);
 #else
     // RISC-V and other architectures use mutex fallback
-    BAIDU_SCOPED_LOCK(_mutex);
+    BAIDU_SCOPED_LOCK(const_cast<FastPthreadMutex&>(_mutex));
     _value = value;
 #endif // __x86_64__ || __ARM_NEON
 }