Merge pull request numpy#121 from howjmay/vfms_f32

howjmay · web-flow · commit 909037dec710 · 2023-11-18T15:19:15.000+01:00
feat: Add vfms_f32
diff --git a/neon2rvv.h b/neon2rvv.h
@@ -580,7 +580,9 @@ FORCE_INLINE float32x2_t vfma_f32(float32x2_t __a, float32x2_t __b, float32x2_t
 
 // FORCE_INLINE float32x4_t vfmaq_f32(float32x4_t __a, float32x4_t __b, float32x4_t __c);
 
-// FORCE_INLINE float32x2_t vfms_f32(float32x2_t __a, float32x2_t __b, float32x2_t __c);
+FORCE_INLINE float32x2_t vfms_f32(float32x2_t __a, float32x2_t __b, float32x2_t __c) {
+  return __riscv_vfnmsac_vv_f32m1(__a, __b, __c, 2);
+}
 
 // FORCE_INLINE float32x4_t vfmsq_f32(float32x4_t __a, float32x4_t __b, float32x4_t __c);
 
diff --git a/tests/impl.cpp b/tests/impl.cpp
@@ -1360,7 +1360,21 @@ result_t test_vfma_f32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
 
 result_t test_vfmaq_f32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
 
-result_t test_vfms_f32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
+result_t test_vfms_f32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
+  const float *_a = (float *)impl.test_cases_float_pointer1;
+  const float *_b = (float *)impl.test_cases_float_pointer2;
+  const float *_c = (float *)impl.test_cases_float_pointer3;
+  float _d[2];
+  for (int i = 0; i < 2; i++) {
+    _d[i] = _a[i] - _b[i] * _c[i];
+  }
+
+  float32x2_t a = vld1_f32(_a);
+  float32x2_t b = vld1_f32(_b);
+  float32x2_t c = vld1_f32(_c);
+  float32x2_t d = vfms_f32(a, b, c);
+  return validate_float_error(d, _d[0], _d[1], 0.001f);
+}
 
 result_t test_vfmsq_f32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
 
