25 #ifdef ARM_COMPUTE_ENABLE_SVE
27 #include "../std_transforms_sve.hpp"
28 #include "../performance_parameters.hpp"
31 const __fp16 *, const __fp16 *, \
32 __fp16 *, int, int, int
37 void sve_interleaved_fp16_mla_8x3VL( ARGLIST );
38 void sve_interleaved_fp16_mla_8x3VL_a64fx( ARGLIST );
40 class cls_sve_interleaved_fp16_mla_8x3VL
43 typedef __fp16 operand_type;
44 typedef __fp16 result_type;
46 typedef void (*kern_type)( ARGLIST );
49 static constexpr
unsigned int out_height()
54 static unsigned int out_width()
56 return get_vector_length<__fp16>() * 3;
59 static constexpr
unsigned int k_unroll()
65 StdTransformsSVE<operand_type, result_type, 8, 3, 1, 1> transforms = {};
66 StdTransformsSVE<operand_type, result_type, 8, 3, 1, 1, true> transforms_quantized = {};
68 static inline PerformanceParameters get_performance_parameters(
const CPUInfo *
ci)
71 if (std::is_same<T, __fp16>::value) {
74 return { 15.96, 3.85, 6.91 };
76 return { 13.84, 2.07, 2.52 };
78 return { 31.90, 5.15, 10.34 };
80 return { 44.34, 3.23, 7.06 };
88 kern_type kernel=sve_interleaved_fp16_mla_8x3VL;
89 cls_sve_interleaved_fp16_mla_8x3VL(
const CPUInfo *
ci)
95 kernel=sve_interleaved_fp16_mla_8x3VL_a64fx;
104 #endif // ARM_COMPUTE_ENABLE_SVE