Draft: Fused-multiply-add vectorization

src/pystencils/backends/arm_instruction_sets.py

@@ -38,6 +38,8 @@ def get_vector_instruction_set_arm(data_type='double', instruction_set='neon'):
         '-': 'sub[0, 1]',
         '*': 'mul[0, 1]',
         '/': 'div[0, 1]',
+        '*+': 'fma[2, 1, 0]',  # 2*1+0
+        '-*+': 'fms[2, 1, 0]',  # -(2*1+0)
         'sqrt': 'sqrt[0]',
 
         'loadU': 'ld1[0]',
@@ -68,6 +70,10 @@ def get_vector_instruction_set_arm(data_type='double', instruction_set='neon'):
         result['bytes'] = bitwidth // 8
     if instruction_set.startswith('sve') or instruction_set == 'sme':
         base_names['stream'] = 'stnt1[0, 1]'
+        base_names['*+'] = 'mad[2, 1, 0]'
+        base_names['*-'] = 'msb[2, 1, 0]'
+        base_names['-*+'] = 'nmad[2, 1, 0]'
+        base_names['-*-'] = 'nmsb[2, 1, 0]'
         prefix = 'sv'
         suffix = f'_f{bits[data_type]}'
     elif instruction_set == 'neon':