Update AVX512 Detection (#2621)

* Update cpuinfo AVX512 detection

* Missing conversion from `_mm256` to `_mm256i`
Co-authored-by: NOlatunji Ruwase <olruwase@microsoft.com>

csrc/includes/simd.h

@@ -21,7 +21,8 @@
 #define SIMD_DIV(x, y) _mm512_div_ps(x, y)
 #define SIMD_WIDTH 16
 
-#define SIMD_LOAD2(x, h) ((h) ? _mm512_cvtph_ps(_mm256_loadu_ps(x)) : _mm512_loadu_ps(x))
+#define SIMD_LOAD2(x, h) \
+    ((h) ? _mm512_cvtph_ps(_mm256_castps_si256(_mm256_loadu_ps(x))) : _mm512_loadu_ps(x))
 #define SIMD_STORE2(x, d, h)                                                                      \
     ((h) ? _mm256_store_ps(x, _mm256_castsi256_ps(_mm512_cvtps_ph(d, _MM_FROUND_TO_NEAREST_INT))) \
          : _mm512_storeu_ps(x, d))

op_builder/builder.py

@@ -376,6 +376,8 @@ class OpBuilder(ABC):
             cpu_info['arch'] = 'X86_64'
             if 'avx512' in result:
                 cpu_info['flags'] += 'avx512,'
+            elif 'avx512f' in result:
+                cpu_info['flags'] += 'avx512f,'
             if 'avx2' in result:
                 cpu_info['flags'] += 'avx2'
         elif 'ppc64le' in result:
@@ -402,7 +404,7 @@ class OpBuilder(ABC):
                 return '-D__SCALAR__'
 
         if cpu_info['arch'] == 'X86_64':
-            if 'avx512' in cpu_info['flags']:
+            if 'avx512' in cpu_info['flags'] or 'avx512f' in cpu_info['flags']:
                 return '-D__AVX512__'
             elif 'avx2' in cpu_info['flags']:
                 return '-D__AVX256__'