feat: Implement _MM_GET_ROUNDING_MODE

sse2neon.h

@@ -6238,6 +6238,7 @@ FORCE_INLINE __m128d _mm_blendv_pd(__m128d _a, __m128d _b, __m128d _mask)
 #endif
 }
 
+// The bit field mapping to the FPCR(floating-point control register)
 typedef struct {
     uint16_t res0;
     uint8_t res1 : 6;
@@ -6249,6 +6250,34 @@ typedef struct {
 #endif
 } fpcr_bitfield;
 
+// Macro: Get the rounding mode bits from the MXCSR control and status register.
+// The rounding mode may contain any of the following flags: _MM_ROUND_NEAREST,
+// _MM_ROUND_DOWN, _MM_ROUND_UP, _MM_ROUND_TOWARD_ZERO
+// https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_MM_GET_ROUNDING_MODE
+FORCE_INLINE unsigned int _MM_GET_ROUNDING_MODE()
+{
+    union {
+        fpcr_bitfield field;
+#if defined(__aarch64__)
+        uint64_t value;
+#else
+        uint32_t value;
+#endif
+    } r;
+
+#if defined(__aarch64__)
+    asm volatile("mrs %0, FPCR" : "=r"(r.value)); /* read */
+#else
+    asm volatile("vmrs %0, FPSCR" : "=r"(r.value)); /* read */
+#endif
+
+    if (r.field.bit22) {
+        return r.field.bit23 ? _MM_ROUND_TOWARD_ZERO : _MM_ROUND_UP;
+    } else {
+        return r.field.bit23 ? _MM_ROUND_DOWN : _MM_ROUND_NEAREST;
+    }
+}
+
 // Macro: Set the rounding mode bits of the MXCSR control and status register to
 // the value in unsigned 32-bit integer a. The rounding mode may contain any of
 // the following flags: _MM_ROUND_NEAREST, _MM_ROUND_DOWN, _MM_ROUND_UP,

tests/impl.cpp

@@ -1688,6 +1688,25 @@ result_t test_mm_free(const SSE2NEONTestImpl &impl, uint32_t i)
     return TEST_UNIMPL;
 }
 
+result_t test_mm_get_rounding_mode(const SSE2NEONTestImpl &impl, uint32_t i)
+{
+    int res_toward_zero, res_to_neg_inf, res_to_pos_inf, res_nearest;
+    _MM_SET_ROUNDING_MODE(_MM_ROUND_TOWARD_ZERO);
+    res_toward_zero = _MM_GET_ROUNDING_MODE() == _MM_ROUND_TOWARD_ZERO ? 1 : 0;
+    _MM_SET_ROUNDING_MODE(_MM_ROUND_DOWN);
+    res_to_neg_inf = _MM_GET_ROUNDING_MODE() == _MM_ROUND_DOWN ? 1 : 0;
+    _MM_SET_ROUNDING_MODE(_MM_ROUND_UP);
+    res_to_pos_inf = _MM_GET_ROUNDING_MODE() == _MM_ROUND_UP ? 1 : 0;
+    _MM_SET_ROUNDING_MODE(_MM_ROUND_NEAREST);
+    res_nearest = _MM_GET_ROUNDING_MODE() == _MM_ROUND_NEAREST ? 1 : 0;
+
+    if (res_toward_zero && res_to_neg_inf && res_to_pos_inf && res_nearest) {
+        return TEST_SUCCESS;
+    } else {
+        return TEST_FAIL;
+    }
+}
+
 result_t test_mm_getcsr(const SSE2NEONTestImpl &impl, uint32_t i)
 {
     return TEST_UNIMPL;

tests/impl.h

@@ -72,6 +72,7 @@
     TYPE(mm_div_ss)                \
     TYPE(mm_extract_pi16)          \
     TYPE(mm_free)                  \
+    TYPE(mm_get_rounding_mode)     \
     TYPE(mm_getcsr)                \
     TYPE(mm_insert_pi16)           \
     TYPE(mm_load_ps)               \