feat: Add vrecps[s|d]_[f32|f64]

howjmay · Jul 30, 2024 · f9dd0f0 · f9dd0f0
1 parent 1d79156
commit f9dd0f0
Show file tree

Hide file tree

Showing 3 changed files with 44 additions and 22 deletions.
diff --git a/neon2rvv.h b/neon2rvv.h
@@ -4796,9 +4796,9 @@ FORCE_INLINE float64x2_t vrecpsq_f64(float64x2_t a, float64x2_t b) {
   return __riscv_vfnmsac_vv_f64m1(vdupq_n_f64(2.0), a, b, 2);
 }
 
-// FORCE_INLINE float32_t vrecpss_f32(float32_t a, float32_t b);
+FORCE_INLINE float32_t vrecpss_f32(float32_t a, float32_t b) { return 2.0 - a * b; }
 
-// FORCE_INLINE float64_t vrecpsd_f64(float64_t a, float64_t b);
+FORCE_INLINE float64_t vrecpsd_f64(float64_t a, float64_t b) { return 2.0 - a * b; }
 
 // FORCE_INLINE float32x2_t vsqrt_f32(float32x2_t a);
 
@@ -5650,21 +5650,21 @@ FORCE_INLINE uint32x4_t vqrshlq_u32(uint32x4_t a, int32x4_t b) {
 
 // FORCE_INLINE uint64x2_t vqrshlq_u64(uint64x2_t a, int64x2_t b);
 
-// FORCE_INLINE int8_t vqrshlb_s8(int8_t a, int8_t b);
+FORCE_INLINE int8_t vqrshlb_s8(int8_t a, int8_t b);
 
-// FORCE_INLINE int16_t vqrshlh_s16(int16_t a, int16_t b);
+FORCE_INLINE int16_t vqrshlh_s16(int16_t a, int16_t b);
 
-// FORCE_INLINE int32_t vqrshls_s32(int32_t a, int32_t b);
+FORCE_INLINE int32_t vqrshls_s32(int32_t a, int32_t b);
 
-// FORCE_INLINE int64_t vqrshld_s64(int64_t a, int64_t b);
+FORCE_INLINE int64_t vqrshld_s64(int64_t a, int64_t b);
 
-// FORCE_INLINE uint8_t vqrshlb_u8(uint8_t a, int8_t b);
+FORCE_INLINE uint8_t vqrshlb_u8(uint8_t a, int8_t b);
 
-// FORCE_INLINE uint16_t vqrshlh_u16(uint16_t a, int16_t b);
+FORCE_INLINE uint16_t vqrshlh_u16(uint16_t a, int16_t b);
 
-// FORCE_INLINE uint32_t vqrshls_u32(uint32_t a, int32_t b);
+FORCE_INLINE uint32_t vqrshls_u32(uint32_t a, int32_t b);
 
-// FORCE_INLINE uint64_t vqrshld_u64(uint64_t a, int64_t b);
+FORCE_INLINE uint64_t vqrshld_u64(uint64_t a, int64_t b);
 
 FORCE_INLINE int8x8_t vshr_n_s8(int8x8_t a, const int b) {
   const int imm = b - (b >> 3);

diff --git a/tests/impl.cpp b/tests/impl.cpp
@@ -16984,9 +16984,31 @@ result_t test_vrecpsq_f64(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
 #endif  // ENABLE_TEST_ALL
 }
 
-result_t test_vrecpss_f32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
+result_t test_vrecpss_f32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
+#ifdef ENABLE_TEST_ALL
+  const float *_a = (const float *)impl.test_cases_float_pointer1;
+  const float *_b = (const float *)impl.test_cases_float_pointer2;
+  float _c, c;
+  _c = 2.0 - _a[0] * _b[0];
+  c = vrecpss_f32(_a[0], _b[0]);
+  return validate_float_error(c, _c, 0.001f);
+#else
+  return TEST_UNIMPL;
+#endif  // ENABLE_TEST_ALL
+}
 
-result_t test_vrecpsd_f64(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
+result_t test_vrecpsd_f64(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
+#ifdef ENABLE_TEST_ALL
+  const double *_a = (const double *)impl.test_cases_float_pointer1;
+  const double *_b = (const double *)impl.test_cases_float_pointer2;
+  double _c, c;
+  _c = 2.0 - _a[0] * _b[0];
+  c = vrecpsd_f64(_a[0], _b[0]);
+  return validate_double_error(c, _c, 0.001f);
+#else
+  return TEST_UNIMPL;
+#endif  // ENABLE_TEST_ALL
+}
 
 result_t test_vsqrt_f32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
 

diff --git a/tests/impl.h b/tests/impl.h
@@ -959,8 +959,8 @@
   _(vrecpsq_f32)                                                                 \
   _(vrecps_f64)                                                                  \
   _(vrecpsq_f64)                                                                 \
-  /*_(vrecpss_f32)                                                            */ \
-  /*_(vrecpsd_f64)                                                            */ \
+  _(vrecpss_f32)                                                                 \
+  _(vrecpsd_f64)                                                                 \
   /*_(vsqrt_f32)                                                              */ \
   /*_(vsqrtq_f32)                                                             */ \
   /*_(vsqrt_f64)                                                              */ \
@@ -1047,14 +1047,14 @@
   _(vqrshlq_u16)                                                                 \
   _(vqrshlq_u32)                                                                 \
   _(vqrshlq_u64)                                                                 \
-  /*_(vqrshlb_s8)                                                             */ \
-  /*_(vqrshlh_s16)                                                            */ \
-  /*_(vqrshls_s32)                                                            */ \
-  /*_(vqrshld_s64)                                                            */ \
-  /*_(vqrshlb_u8)                                                             */ \
-  /*_(vqrshlh_u16)                                                            */ \
-  /*_(vqrshls_u32)                                                            */ \
-  /*_(vqrshld_u64)                                                            */ \
+  _(vqrshlb_s8)                                                              \
+  _(vqrshlh_s16)                                                             \
+  _(vqrshls_s32)                                                             \
+  _(vqrshld_s64)                                                             \
+  _(vqrshlb_u8)                                                              \
+  _(vqrshlh_u16)                                                             \
+  _(vqrshls_u32)                                                             \
+  _(vqrshld_u64)                                                             \
   _(vshr_n_s8)                                                                   \
   _(vshr_n_s16)                                                                  \
   _(vshr_n_s32)                                                                  \