[libclc][amdgpu] Implement native_exp via builtin

This came up during a discussion on #129679, which has been split out as
a preparatory commit.

An example of the AMDGPU codegen is:

    define <2 x float> @_Z10native_expDv2_f(<2 x float> %x) {
      %0 = extractelement <2 x float> %x, i64 0
      %mul.i4 = fmul afn float %0, 0x3FF7154760000000
      %1 = tail call afn float @llvm.amdgcn.exp2.f32(float %mul.i4)
      %vecinit = insertelement <2 x float> poison, float %1, i64 0
      %2 = extractelement <2 x float> %x, i64 1
      %mul.i = fmul afn float %2, 0x3FF7154760000000
      %3 = tail call afn float @llvm.amdgcn.exp2.f32(float %mul.i)
      %vecinit2 = insertelement <2 x float> %vecinit, float %3, i64 1
      ret <2 x float> %vecinit2
    }

Author: frasercrmck

libclc/amdgpu/lib/math/native_exp.cl

@@ -7,7 +7,10 @@
 //===----------------------------------------------------------------------===//
 
 #include <clc/clc.h>
+#include <clc/clcmacro.h>
 
-#define __CLC_BODY <native_exp.inc>
-#define __FLOAT_ONLY
-#include <clc/math/gentype.inc>
+_CLC_OVERLOAD _CLC_DEF float native_exp(float val) {
+  return __builtin_amdgcn_exp2f(val * M_LOG2E_F);
+}
+
+_CLC_UNARY_VECTORIZE(_CLC_OVERLOAD _CLC_DEF, float, native_exp, float)