26 #if defined(__ARM_FEATURE_SVE2) 27 inline svuint8_t svmla_qasymm8_z(svbool_t pg, svuint8_t vd, svfloat32_t vs, svfloat32_t vo)
30 auto vd_low_u16 = svmovlb_u16(vd);
31 auto vd_high_u16 = svmovlt_u16(vd);
34 auto A_u32 = svmovlb_u32(vd_low_u16);
35 auto B_u32 = svmovlt_u32(vd_low_u16);
36 auto C_u32 = svmovlb_u32(vd_high_u16);
37 auto D_u32 = svmovlt_u32(vd_high_u16);
40 auto A_f32 = svcvt_f32_u32_z(pg, A_u32);
41 auto B_f32 = svcvt_f32_u32_z(pg, B_u32);
42 auto C_f32 = svcvt_f32_u32_z(pg, C_u32);
43 auto D_f32 = svcvt_f32_u32_z(pg, D_u32);
46 A_f32 = svmla_f32_z(pg, vo, A_f32, vs);
47 B_f32 = svmla_f32_z(pg, vo, B_f32, vs);
48 C_f32 = svmla_f32_z(pg, vo, C_f32, vs);
49 D_f32 = svmla_f32_z(pg, vo, D_f32, vs);
52 A_u32 = svcvt_u32_f32_z(pg, A_f32);
53 B_u32 = svcvt_u32_f32_z(pg, B_f32);
54 C_u32 = svcvt_u32_f32_z(pg, C_f32);
55 D_u32 = svcvt_u32_f32_z(pg, D_f32);
58 vd_low_u16 = svqxtnt_u32(svqxtnb_u32(A_u32), B_u32);
59 vd_high_u16 = svqxtnt_u32(svqxtnb_u32(C_u32), D_u32);
62 const auto res = svqxtnt_u16(svqxtnb_u16(vd_low_u16), vd_high_u16);
66 inline svint8_t svmla_qasymm8_signed_z(svbool_t pg, svint8_t vd, svfloat32_t vs, svfloat32_t vo)
69 auto vd_low_s16 = svmovlb_s16(vd);
70 auto vd_high_s16 = svmovlt_s16(vd);
73 auto A_s32 = svmovlb_s32(vd_low_s16);
74 auto B_s32 = svmovlt_s32(vd_low_s16);
75 auto C_s32 = svmovlb_s32(vd_high_s16);
76 auto D_s32 = svmovlt_s32(vd_high_s16);
79 auto A_f32 = svcvt_f32_s32_z(pg, A_s32);
80 auto B_f32 = svcvt_f32_s32_z(pg, B_s32);
81 auto C_f32 = svcvt_f32_s32_z(pg, C_s32);
82 auto D_f32 = svcvt_f32_s32_z(pg, D_s32);
85 A_f32 = svmla_f32_z(pg, vo, A_f32, vs);
86 B_f32 = svmla_f32_z(pg, vo, B_f32, vs);
87 C_f32 = svmla_f32_z(pg, vo, C_f32, vs);
88 D_f32 = svmla_f32_z(pg, vo, D_f32, vs);
91 A_s32 = svcvt_s32_f32_z(pg, A_f32);
92 B_s32 = svcvt_s32_f32_z(pg, B_f32);
93 C_s32 = svcvt_s32_f32_z(pg, C_f32);
94 D_s32 = svcvt_s32_f32_z(pg, D_f32);
97 vd_low_s16 = svqxtnt_s32(svqxtnb_s32(A_s32), B_s32);
98 vd_high_s16 = svqxtnt_s32(svqxtnb_s32(C_s32), D_s32);
101 const auto res = svqxtnt_s16(svqxtnb_s16(vd_low_s16), vd_high_s16);
Copyright (c) 2017-2021 Arm Limited.