arch: support grace with gcc,arm and nvc

Author: mloubout

devito/arch/archinfo.py

@@ -29,7 +29,7 @@
            # ARM CPUs
            'ARM', 'AppleArm', 'M1', 'M2', 'M3',
            'Graviton', 'GRAVITON2', 'GRAVITON3', 'GRAVITON4',
-           'Cortex',
+           'Cortex', 'NvidiaArm', 'GRACE',
            # Other legacy CPUs
            'POWER8', 'POWER9',
            # Generic GPUs
@@ -811,6 +811,16 @@ def mtune(self):
         return f'cortex-a{self.version}'
 
 
+class NvidiaArm(Arm):
+
+    @cached_property
+    def march(self):
+        if self.name == 'grace':
+            return 'neoverse-v2'
+        else:
+            return 'native'
+
+
 class Amd(Cpu64):
 
     known_isas = ('cpp', 'sse', 'avx', 'avx2')
@@ -1027,6 +1037,7 @@ def march(cls):
 M3 = AppleArm('m3')
 CORTEX = Cortex('cortex')
 CORTEXA76 = Cortex('cortexa76')
+GRACE = NvidiaArm('grace')
 
 AMD = Amd('amd')
 

devito/arch/compiler.py

@@ -14,7 +14,7 @@
                               call_capture_output as _call_capture_output)
 
 from devito.arch import (AMDGPUX, Cpu64, AppleArm, NvidiaDevice, POWER8, POWER9,
-                         Graviton, Cortex, IntelDevice, get_nvidia_cc,
+                         Graviton, Cortex, IntelDevice, get_nvidia_cc, NvidiaArm,
                          check_cuda_runtime, get_m1_llvm_path)
 from devito.exceptions import CompilationError
 from devito.logger import debug, warning
@@ -439,7 +439,7 @@ def __init_finalize__(self, **kwargs):
         if platform in [POWER8, POWER9]:
             # -march isn't supported on power architectures, is -mtune needed?
             self.cflags = ['-mcpu=native'] + self.cflags
-        elif isinstance(platform, Graviton):
+        elif isinstance(platform, (Graviton, NvidiaArm)):
             self.cflags = [f'-mcpu={platform.march}'] + self.cflags
         elif isinstance(platform, Cortex):
             self.cflags += [f'-march={platform.march}']
@@ -614,6 +614,8 @@ def __init_finalize__(self, **kwargs):
         elif isinstance(platform, Cpu64):
             if language == 'openmp':
                 self.cflags.append('-mp')
+            if isinstance(platform, NvidiaArm):
+                self.cflags.append(f'-mcpu={platform.march}')
 
         if not configuration['safe-math']:
             self.cflags.append('-fast')