[CUDA] Remove htanh from unsupported math ops for CUDA 12.8

This PR removes htanh from the list of unsupported CUDA half
operators, as it is started to be supported since CUDA 12.8.

A better option is to check the CUDA version at compile time,
while due to the rare use of htanh, the removal seems a simpler
way to go with.

Author: MasterJH5574

src/target/source/literal/cuda_half_t.h

@@ -317,7 +317,6 @@ static inline __device__ __host__ half HALF_MATH_NAME(half x) {          \
 #if defined(__CUDA_ARCH__)
 #if (__CUDA_ARCH__ >= 530)
 CUDA_UNSUPPORTED_HALF_MATH_BINARY(hpow, powf)
-CUDA_UNSUPPORTED_HALF_MATH_UNARY(htanh, tanhf)
 CUDA_UNSUPPORTED_HALF_MATH_UNARY(htan, tanf)
 CUDA_UNSUPPORTED_HALF_MATH_UNARY(hatan, atanf)
 CUDA_UNSUPPORTED_HALF_MATH_UNARY(herf, erf)
@@ -358,7 +357,6 @@ static inline __device__ __host__ nv_bfloat16 HALF_MATH_NAME(nv_bfloat16 x) {
 }
 
 CUDA_UNSUPPORTED_HALF_MATH_BINARY(hpow, powf)
-CUDA_UNSUPPORTED_HALF_MATH_UNARY(htanh, tanhf)
 CUDA_UNSUPPORTED_HALF_MATH_UNARY(htan, tanf)
 CUDA_UNSUPPORTED_HALF_MATH_UNARY(hatan, atanf)
 CUDA_UNSUPPORTED_HALF_MATH_UNARY(herf, erf)

tests/python/meta_schedule/test_meta_schedule_mma_m16n8k8_auto_tensorization.py

@@ -717,7 +717,6 @@ class TVM_ALIGNED(2) half {
 #if defined(__CUDA_ARCH__)
 #if (__CUDA_ARCH__ >= 530)
 CUDA_UNSUPPORTED_HALF_MATH_BINARY(hpow, powf)
-CUDA_UNSUPPORTED_HALF_MATH_UNARY(htanh, tanhf)
 CUDA_UNSUPPORTED_HALF_MATH_UNARY(htan, tanf)
 CUDA_UNSUPPORTED_HALF_MATH_UNARY(hatan, atanf)
 CUDA_UNSUPPORTED_HALF_MATH_UNARY(herf, erf)