feat: Add vaddl_s8

howjmay · Oct 17, 2023 · 29e2d13 · 29e2d13
1 parent 741651f
commit 29e2d13
Show file tree

Hide file tree

Showing 2 changed files with 17 additions and 2 deletions.
diff --git a/neon2rvv.h b/neon2rvv.h
@@ -173,7 +173,9 @@ FORCE_INLINE uint32x4_t vaddq_u32(uint32x4_t __a, uint32x4_t __b) { return __ris
 
 FORCE_INLINE uint64x2_t vaddq_u64(uint64x2_t __a, uint64x2_t __b) { return __riscv_vadd_vv_u64m1(__a, __b, 2); }
 
-// FORCE_INLINE int16x8_t vaddl_s8(int8x8_t __a, int8x8_t __b);
+FORCE_INLINE int16x8_t vaddl_s8(int8x8_t __a, int8x8_t __b) {
+  return __riscv_vget_v_i16m2_i16m1(__riscv_vwadd_vv_i16m2(__a, __b, 8), 0);
+}
 
 // FORCE_INLINE int32x4_t vaddl_s16(int16x4_t __a, int16x4_t __b);
 

diff --git a/tests/impl.cpp b/tests/impl.cpp
@@ -578,7 +578,20 @@ result_t test_vaddq_u64(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
   return validate_uint64(c, d0, d1);
 }
 
-result_t test_vaddl_s8(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
+result_t test_vaddl_s8(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
+  const int8_t *_a = (int8_t *)impl.test_cases_int_pointer1;
+  const int8_t *_b = (int8_t *)impl.test_cases_int_pointer1;
+  int16_t _c[8];
+
+  for (int i = 0; i < 8; i++) {
+    _c[i] = (int16_t)(_a[i] + _b[i]);
+  }
+
+  int8x8_t a = vld1_s8(_a);
+  int8x8_t b = vld1_s8(_b);
+  int16x8_t c = vaddl_s8(a, b);
+  return validate_int16(c, _c[0], _c[1], _c[2], _c[3], _c[4], _c[5], _c[6], _c[7]);
+}
 
 result_t test_vaddl_s16(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }