Merge pull request numpy#79 from howjmay/vmovn_s16

howjmay · web-flow · commit fa90b59f03fd · 2023-11-12T19:45:51.000+01:00
feat: Add vmovn_s16
diff --git a/neon2rvv.h b/neon2rvv.h
@@ -2100,7 +2100,11 @@ FORCE_INLINE int8x8_t vget_low_s8(int8x16_t __a) {
 
 // FORCE_INLINE uint32x4_t vcvtq_n_u32_f32(float32x4_t __a, const int __b);
 
-// FORCE_INLINE int8x8_t vmovn_s16(int16x8_t __a);
+FORCE_INLINE int8x8_t vmovn_s16(int16x8_t __a) {
+  uint8_t mask_arr[] = {85, 85};
+  vbool8_t mask = __riscv_vlm_v_b8(mask_arr, 16);
+  return __riscv_vcompress_vm_i8m1(__riscv_vreinterpret_v_i16m1_i8m1(__a), mask, 16);
+}
 
 // FORCE_INLINE int16x4_t vmovn_s32(int32x4_t __a);
 
diff --git a/tests/impl.cpp b/tests/impl.cpp
@@ -3219,7 +3219,16 @@ result_t test_vcvtq_n_f32_u32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { r
 
 result_t test_vcvtq_n_u32_f32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
 
-result_t test_vmovn_s16(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
+result_t test_vmovn_s16(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) {
+  const int16_t *_a = (int16_t *)impl.test_cases_int_pointer1;
+  int8_t _c[8];
+  for (int i = 0; i < 8; i++) {
+    _c[i] = _a[i] & 0xff;
+  }
+  int16x8_t a = vld1q_s16(_a);
+  int8x8_t c = vmovn_s16(a);
+  return validate_int8(c, _c[0], _c[1], _c[2], _c[3], _c[4], _c[5], _c[6], _c[7]);
+}
 
 result_t test_vmovn_s32(const NEON2RVV_TEST_IMPL &impl, uint32_t iter) { return TEST_UNIMPL; }
 
