Merge pull request #495 from howjmay/vqneg

feat: Add vqneg[b|h|s|d]_[s8|s16|s32|s64]
howjmay · Jul 31, 2024 · 519b350 · 519b350
2 parents a34f81f + a953e83
commit 519b350
Show file tree

Hide file tree

Showing 3 changed files with 92 additions and 12 deletions.
diff --git a/neon2rvv.h b/neon2rvv.h
@@ -7197,13 +7197,37 @@ FORCE_INLINE int64x2_t vqnegq_s64(int64x2_t a) {
   return __riscv_vmerge_vxm_i64m1(a_neg, INT64_MAX, min_mask, 2);
 }
 
-// FORCE_INLINE int8_t vqnegb_s8(int8_t a);
+FORCE_INLINE int8_t vqnegb_s8(int8_t a) {
+  if (a == INT8_MIN) {
+    return INT8_MAX;
+  } else {
+    return -a;
+  }
+}
 
-// FORCE_INLINE int16_t vqnegh_s16(int16_t a);
+FORCE_INLINE int16_t vqnegh_s16(int16_t a) {
+  if (a == INT16_MIN) {
+    return INT16_MAX;
+  } else {
+    return -a;
+  }
+}
 
-// FORCE_INLINE int32_t vqnegs_s32(int32_t a);
+FORCE_INLINE int32_t vqnegs_s32(int32_t a) {
+  if (a == INT32_MIN) {
+    return INT32_MAX;
+  } else {
+    return -a;
+  }
+}
 
-// FORCE_INLINE int64_t vqnegd_s64(int64_t a);
+FORCE_INLINE int64_t vqnegd_s64(int64_t a) {
+  if (a == INT64_MIN) {
+    return INT64_MAX;
+  } else {
+    return -a;
+  }
+}
 
 FORCE_INLINE int8x8_t vmvn_s8(int8x8_t a) { return __riscv_vnot_v_i8m1(a, 8); }
 

diff --git a/tests/impl.cpp b/tests/impl.cpp
@@ -25927,13 +25927,69 @@ result_t test_vqnegq_s64(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
 #endif  // ENABLE_TEST_ALL
 }
 
-result_t test_vqnegb_s8(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
+result_t test_vqnegb_s8(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
+#ifdef ENABLE_TEST_ALL
+  int8_t *_a = (int8_t *)impl.test_cases_int_pointer1;
+  int8_t _c, c;
+  if (_a[0] == INT8_MIN) {
+    _c = INT8_MAX;
+  } else {
+    _c = -_a[0];
+  }
+  c = vqnegb_s8(_a[0]);
+  return c == _c ? TEST_SUCCESS : TEST_FAIL;
+#else
+  return TEST_UNIMPL;
+#endif  // ENABLE_TEST_ALL
+}
 
-result_t test_vqnegh_s16(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
+result_t test_vqnegh_s16(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
+#ifdef ENABLE_TEST_ALL
+  int16_t *_a = (int16_t *)impl.test_cases_int_pointer1;
+  int16_t _c, c;
+  if (_a[0] == INT16_MIN) {
+    _c = INT16_MAX;
+  } else {
+    _c = -_a[0];
+  }
+  c = vqnegh_s16(_a[0]);
+  return c == _c ? TEST_SUCCESS : TEST_FAIL;
+#else
+  return TEST_UNIMPL;
+#endif  // ENABLE_TEST_ALL
+}
 
-result_t test_vqnegs_s32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
+result_t test_vqnegs_s32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
+#ifdef ENABLE_TEST_ALL
+  int32_t *_a = (int32_t *)impl.test_cases_int_pointer1;
+  int32_t _c, c;
+  if (_a[0] == INT32_MIN) {
+    _c = INT32_MAX;
+  } else {
+    _c = -_a[0];
+  }
+  c = vqnegs_s32(_a[0]);
+  return c == _c ? TEST_SUCCESS : TEST_FAIL;
+#else
+  return TEST_UNIMPL;
+#endif  // ENABLE_TEST_ALL
+}
 
-result_t test_vqnegd_s64(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
+result_t test_vqnegd_s64(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
+#ifdef ENABLE_TEST_ALL
+  int64_t *_a = (int64_t *)impl.test_cases_int_pointer1;
+  int64_t _c, c;
+  if (_a[0] == INT64_MIN) {
+    _c = INT64_MAX;
+  } else {
+    _c = -_a[0];
+  }
+  c = vqnegd_s64(_a[0]);
+  return c == _c ? TEST_SUCCESS : TEST_FAIL;
+#else
+  return TEST_UNIMPL;
+#endif  // ENABLE_TEST_ALL
+}
 
 result_t test_vmvn_s8(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
 #ifdef ENABLE_TEST_ALL

diff --git a/tests/impl.h b/tests/impl.h
@@ -1340,10 +1340,10 @@
   _(vqnegq_s32)                                                                  \
   _(vqneg_s64)                                                                   \
   _(vqnegq_s64)                                                                  \
-  /*_(vqnegb_s8)                                                              */ \
-  /*_(vqnegh_s16)                                                             */ \
-  /*_(vqnegs_s32)                                                             */ \
-  /*_(vqnegd_s64)                                                             */ \
+  _(vqnegb_s8)                                                                   \
+  _(vqnegh_s16)                                                                  \
+  _(vqnegs_s32)                                                                  \
+  _(vqnegd_s64)                                                                  \
   _(vmvn_s8)                                                                     \
   _(vmvn_s16)                                                                    \
   _(vmvn_s32)                                                                    \