Vectorization improvements

After we cleaned up vectorization support as part of our ARM Neon experiments a few weeks ago (!188 (merged), !220 (merged), !222 (merged)), I did the same thing with AltiVec/VSX intrinsics for POWER processors. Adding a new SIMD instruction set to pystencils really is just a matter of some quick find-and-replace now. I had test access to a POWER8 machine today, ran in both little-endian and big-endian mode, and all tests passed. So pystencils now actually supports all SIMD instruction sets out there (ignoring MIPS and SPARC processors, which are essentially dead).

This pull request also contains some minor unrelated changes:

• switches the AES RNG to aligned stores
• adds a missing pytest.importorskip
• fixes the vec_any/vec_all operations (which used to only work on 256 bit doubles)
• removes the q_registers argument from get_vector_instruction_set because there is no point in using half-width vectors
• fix the AES-NI RNG on Ice Lake/Tiger Lake processors

pystencils/backends/arm_instruction_sets.py

@@ -13,7 +13,7 @@ def get_argument_string(function_shortcut):
     return arg_string
 
 
-def get_vector_instruction_set_arm(data_type='double', instruction_set='neon', q_registers=True):
+def get_vector_instruction_set_arm(data_type='double', instruction_set='neon'):
     if instruction_set != 'neon':
         raise NotImplementedError(instruction_set)
 
@@ -42,16 +42,9 @@ def get_vector_instruction_set_arm(data_type='double', instruction_set='neon', q
             'float': 32,
             'int': 32}
 
-    if q_registers is True:
-        q_reg = 'q'
-        width = 128 // bits[data_type]
-        intwidth = 128 // bits['int']
-        suffix = f'q_f{bits[data_type]}'
-    else:
-        q_reg = ''
-        width = 64 // bits[data_type]
-        intwidth = 64 // bits['int']
-        suffix = f'_f{bits[data_type]}'
+    width = 128 // bits[data_type]
+    intwidth = 128 // bits['int']
+    suffix = f'q_f{bits[data_type]}'
 
     result = dict()
 
@@ -63,10 +56,10 @@ def get_vector_instruction_set_arm(data_type='double', instruction_set='neon', q
 
         result[intrinsic_id] = 'v' + name + suffix + arg_string
 
-    result['makeVecConst'] = f'vdup{q_reg}_n_f{bits[data_type]}' + '({0})'
+    result['makeVecConst'] = f'vdupq_n_f{bits[data_type]}' + '({0})'
     result['makeVec'] = f'makeVec_f{bits[data_type]}' + '(' + ", ".join(['{' + str(i) + '}' for i in range(width)]) + \
         ')'
-    result['makeVecConstInt'] = f'vdup{q_reg}_n_s{bits["int"]}' + '({0})'
+    result['makeVecConstInt'] = f'vdupq_n_s{bits["int"]}' + '({0})'
     result['makeVecInt'] = f'makeVec_s{bits["int"]}' + '({0}, {1}, {2}, {3})'
 
     result['+int'] = f"vaddq_s{bits['int']}" + "({0}, {1})"
@@ -80,11 +73,11 @@ def get_vector_instruction_set_arm(data_type='double', instruction_set='neon', q
     result['bool'] = f'uint{bits[data_type]}x{width}_t'
     result['headers'] = ['<arm_neon.h>', '"arm_neon_helpers.h"']
 
-    result['!='] = f'vmvn{q_reg}_u{bits[data_type]}({result["=="]})'
+    result['!='] = f'vmvnq_u{bits[data_type]}({result["=="]})'
 
-    result['&'] = f'vand{q_reg}_u{bits[data_type]}' + '({0}, {1})'
-    result['|'] = f'vorr{q_reg}_u{bits[data_type]}' + '({0}, {1})'
-    result['blendv'] = f'vbsl{q_reg}_f{bits[data_type]}' + '({2}, {1}, {0})'
+    result['&'] = f'vandq_u{bits[data_type]}' + '({0}, {1})'
+    result['|'] = f'vorrq_u{bits[data_type]}' + '({0}, {1})'
+    result['blendv'] = f'vbslq_f{bits[data_type]}' + '({2}, {1}, {0})'
     result['any'] = f'vaddlvq_u8(vreinterpretq_u8_u{bits[data_type]}({{0}})) > 0'
     result['all'] = f'vaddlvq_u8(vreinterpretq_u8_u{bits[data_type]}({{0}})) == 16*0xff'