Merge pull request #454 from howjmay/vmul_lane_f64

feat: Add vmul[q]_lane_f64

Author: howjmay

neon2rvv.h

@@ -8775,9 +8775,15 @@ FORCE_INLINE float32x4_t vmulq_lane_f32(float32x4_t a, float32x2_t b, const int
   return __riscv_vfmul_vv_f32m1(a, b_dup_lane, 4);
 }
 
-// FORCE_INLINE float64x1_t vmul_lane_f64(float64x1_t a, float64x1_t v, const int lane);
+FORCE_INLINE float64x1_t vmul_lane_f64(float64x1_t a, float64x1_t b, const int lane) {
+  vfloat64m1_t b_dup_lane = __riscv_vrgather_vx_f64m1(b, lane, 1);
+  return __riscv_vfmul_vv_f64m1(a, b_dup_lane, 1);
+}
 
-// FORCE_INLINE float64x2_t vmulq_lane_f64(float64x2_t a, float64x1_t v, const int lane);
+FORCE_INLINE float64x2_t vmulq_lane_f64(float64x2_t a, float64x1_t b, const int lane) {
+  vfloat64m1_t b_dup_lane = __riscv_vrgather_vx_f64m1(b, lane, 2);
+  return __riscv_vfmul_vv_f64m1(a, b_dup_lane, 2);
+}
 
 // FORCE_INLINE float32_t vmuls_lane_f32(float32_t a, float32x2_t v, const int lane);
 

tests/impl.cpp

@@ -31083,9 +31083,47 @@ result_t test_vmulq_lane_f32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
 #endif  // ENABLE_TEST_ALL
 }
 
-result_t test_vmul_lane_f64(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
+result_t test_vmul_lane_f64(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
+#ifdef ENABLE_TEST_ALL
+  const double *_a = (double *)impl.test_cases_float_pointer1;
+  const double *_b = (double *)impl.test_cases_float_pointer2;
+  double _c[2];
+  float64x1_t a = vld1_f64(_a);
+  float64x1_t b = vld1_f64(_b);
+  float64x1_t c;
+
+  for (int i = 0; i < 1; i++) {
+    _c[i] = _a[i] * _b[0];
+  }
+  c = vmul_lane_f64(a, b, 0);
+  return validate_double(c, _c[0]);
+
+  return TEST_SUCCESS;
+
+#else
+  return TEST_UNIMPL;
+#endif  // ENABLE_TEST_ALL
+}
+
+result_t test_vmulq_lane_f64(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
+#ifdef ENABLE_TEST_ALL
+  const double *_a = (double *)impl.test_cases_float_pointer1;
+  const double *_b = (double *)impl.test_cases_float_pointer2;
+  double _c[4];
+  float64x2_t a = vld1q_f64(_a);
+  float64x1_t b = vld1_f64(_b);
+  float64x2_t c;
 
-result_t test_vmulq_lane_f64(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
+  for (int i = 0; i < 2; i++) {
+    _c[i] = _a[i] * _b[0];
+  }
+  c = vmulq_lane_f64(a, b, 0);
+  return validate_double(c, _c[0], _c[1]);
+
+#else
+  return TEST_UNIMPL;
+#endif  // ENABLE_TEST_ALL
+}
 
 result_t test_vmuls_lane_f32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
 

tests/impl.h

@@ -1890,8 +1890,8 @@
   _(vmulq_lane_s16)                                                              \
   _(vmulq_lane_s32)                                                              \
   _(vmulq_lane_f32)                                                              \
-  /*_(vmul_lane_f64)                                                          */ \
-  /*_(vmulq_lane_f64)                                                         */ \
+  _(vmul_lane_f64)                                                               \
+  _(vmulq_lane_f64)                                                              \
   /*_(vmuls_lane_f32)                                                         */ \
   /*_(vmuld_lane_f64)                                                         */ \
   /*_(vmul_laneq_s16)                                                         */ \