[CUDA] added missing __ldg(const signed char *)

[tra]

Summary

Until now we only had variants for 'char' and 'unsigned char'. In C++ 'char' 'unsigned char' and 'signed char' are three different types and we need overloads for all of them.

clang/lib/Headers/__clang_cuda_intrinsics.h

 inline __device__ int __ldg(const int *ptr) { return __nvvm_ldg_i(ptr); }
 inline __device__ long __ldg(const long *ptr) { return __nvvm_ldg_l(ptr); }
 inline __device__ long long __ldg(const long long *ptr) {
   return __nvvm_ldg_ll(ptr);
 }
 inline __device__ unsigned char __ldg(const unsigned char *ptr) {
   return __nvvm_ldg_uc(ptr);
 }
+inline __device__ signed char __ldg(const signed char *ptr) {
+  return __nvvm_ldg_uc((const unsigned char *)ptr);
+}
 inline __device__ unsigned short __ldg(const unsigned short *ptr) {
   return __nvvm_ldg_us(ptr);
 }
 inline __device__ unsigned int __ldg(const unsigned int *ptr) {
   return __nvvm_ldg_ui(ptr);
 }
 inline __device__ unsigned long __ldg(const unsigned long *ptr) {
   return __nvvm_ldg_ul(ptr);