diff options
Diffstat (limited to 'arm_compute/core/NEON/wrapper/intrinsics/cvt.h')
| -rw-r--r-- | arm_compute/core/NEON/wrapper/intrinsics/cvt.h | 19 |
1 files changed, 19 insertions, 0 deletions
diff --git a/arm_compute/core/NEON/wrapper/intrinsics/cvt.h b/arm_compute/core/NEON/wrapper/intrinsics/cvt.h index 1f22e09a11..5ea9a5dedd 100644 --- a/arm_compute/core/NEON/wrapper/intrinsics/cvt.h +++ b/arm_compute/core/NEON/wrapper/intrinsics/cvt.h @@ -56,6 +56,25 @@ vcvt(const float32x4_t &a) return vcvtq_s32_f32(a); } +#if defined(__ARM_FEATURE_BF16_VECTOR_ARITHMETIC) || defined(ARM_COMPUTE_FORCE_BF16) +/** Convert 2x128-bit floating point vectors into 1x128-bit bfloat16 vector + * + * @param[in] inptr Pointer to the input memory to load values from + * @param[in,out] outptr Pointer to the output memory to store values to + */ +inline void vcvt_bf16_f32(const float *inptr, uint16_t *outptr) +{ + __asm __volatile( + "ldp q0, q1, [%[inptr]]\n" + ".inst 0xea16800\n" // BFCVTN v0, v0 + ".inst 0x4ea16820\n" // BFCVTN2 v0, v1 + "str q0, [%[outptr]]\n" + : [inptr] "+r"(inptr) + : [outptr] "r"(outptr) + : "v0", "v1", "memory"); +} +#endif /* defined(__ARM_FEATURE_BF16_VECTOR_ARITHMETIC) || defined(ARM_COMPUTE_FORCE_BF16) */ + } // namespace wrapper } // namespace arm_compute #endif /* ARM_COMPUTE_WRAPPER_CVT_H */ |