25 #ifdef ARM_COMPUTE_ENABLE_SVE
27 #include "../std_transforms_sve.hpp"
28 #include "../performance_parameters.hpp"
31 unsigned int, const unsigned int *, \
32 IndirectInputArg<uint8_t>, \
35 IndirectOutputArg<uint32_t>, \
36 const uint32_t *, Activation, bool
41 void sve_hybrid_u8u32_dot_6x4VL( ARGLIST );
42 void sve_hybrid_u8u32_dot_6x4VL_a64fx( ARGLIST );
44 class cls_sve_hybrid_u8u32_dot_6x4VL
47 typedef uint8_t lhs_operand_type;
48 typedef uint8_t rhs_operand_type;
49 typedef uint32_t result_type;
51 typedef void (*kern_type)( ARGLIST );
54 static constexpr
unsigned int out_height()
59 static unsigned int out_width()
61 return get_vector_length<uint32_t>() * 4;
64 static constexpr
unsigned int k_unroll()
69 static constexpr
bool supports_accumulate()
74 StdTransformsSVE<rhs_operand_type, result_type, 6, 4, 4> transforms = {};
76 static inline PerformanceParameters get_performance_parameters(
const CPUInfo *
ci)
78 if (std::is_same<T, uint32_t>::value) {
91 if (std::is_same<T, uint8_t>::value) {
94 return { 31.59, 15.67, 0.61 };
96 return { 22.75, 3.90, 0.47 };
98 return { 48.09, 16.24, 0.83 };
100 return { 101.62, 3.15, 0.42 };
108 kern_type kernel=sve_hybrid_u8u32_dot_6x4VL;
109 cls_sve_hybrid_u8u32_dot_6x4VL(
const CPUInfo *
ci)
114 case CPUModel::A64FX:
115 kernel=sve_hybrid_u8u32_dot_6x4VL_a64fx;
124 #endif // ARM_COMPUTE_ENABLE_SVE