Does OpenCL support rvv Intrinsic functions？

[zhaosiying12138]

For example，I have a test_rvv_intrinsics.cl file:

#include <riscv_vector.h>

void softmax_fp16()
{
    size_t vlmax = __riscv_vsetvlmax_e16m4();
    _Float16 *tmp_src_data;
    vfloat16m4_t vx_data = __riscv_vle16_v_f16m4(tmp_src_data, vlmax);
}

And I compile it with the following command:

clang --target=riscv64 -march=rv64imafdcv_zfh_zvfh_zve32f -S -c test_rvv_intrinsics.cl -o -

Clang compiles failed and report the following message:

clang: warning: argument unused during compilation: '-c' [-Wunused-command-line-argument]
test_rvv_intrinsics.cl:7:50: error: passing '__generic _Float16 *__private' to parameter of type 'const _Float16 *' changes address space of pointer
    7 |     vfloat16m4_t vx_data = __riscv_vle16_v_f16m4(tmp_src_data, vlmax);
      |                                                  ^~~~~~~~~~~~
test_rvv_intrinsics.cl:7:28: note: passing argument to parameter here
    7 |     vfloat16m4_t vx_data = __riscv_vle16_v_f16m4(tmp_src_data, vlmax);
      |                            ^
1 error generated.

[llvmbot]

@llvm/issue-subscribers-opencl

Author: 流霞祭司曌曦潆 (zhaosiying12138)

For example，I have a test_rvv_intrinsics.cl file: ```c #include <riscv_vector.h>

void softmax_fp16()
{
size_t vlmax = __riscv_vsetvlmax_e16m4();
_Float16 *tmp_src_data;
vfloat16m4_t vx_data = __riscv_vle16_v_f16m4(tmp_src_data, vlmax);
}

And I compile it with the following command:
```shell
clang --target=riscv64 -march=rv64imafdcv_zfh_zvfh_zve32f -S -c test_rvv_intrinsics.cl -o -

Clang compiles failed and report the following message:

clang: warning: argument unused during compilation: '-c' [-Wunused-command-line-argument]
test_rvv_intrinsics.cl:7:50: error: passing '__generic _Float16 *__private' to parameter of type 'const _Float16 *' changes address space of pointer
    7 |     vfloat16m4_t vx_data = __riscv_vle16_v_f16m4(tmp_src_data, vlmax);
      |                                                  ^~~~~~~~~~~~
test_rvv_intrinsics.cl:7:28: note: passing argument to parameter here
    7 |     vfloat16m4_t vx_data = __riscv_vle16_v_f16m4(tmp_src_data, vlmax);
      |                            ^
1 error generated.

[zhaosiying12138]

I did some research about this problem and I recorded it here, https://www.zhihu.com/question/7816930826/answer/63699633587
It seems like a workaround and I haven't done the exhaustive testing like ninja check-clang.