feat(gi): add more gi api

GitOrigin-RevId: e2ae8c0873366bc920eb0ab3398f41f995c34ffd

dnn/src/fallback/general_intrinsic/gi_float.h

@@ -1750,6 +1750,42 @@ GI_FLOAT32_V3_t GiLoadUzipFloat32V3(const float* ptr) {
 #endif
 }
 
+GI_FORCEINLINE
+GI_FLOAT32_V4_t GiLoadUzipFloat32V4(const float* ptr) {
+#if defined(GI_NEON_INTRINSICS)
+    return vld4q_f32(ptr);
+#elif defined(GI_SSE2_INTRINSICS)
+    GI_FLOAT32_V4_t v;
+    __m128 tmp0, tmp1, tmp2, tmp3;
+    v.val[0] = GiLoadFloat32(ptr);
+    v.val[1] = GiLoadFloat32((ptr + 4));
+    v.val[2] = GiLoadFloat32((ptr + 8));
+    v.val[3] = GiLoadFloat32((ptr + 12));
+
+    tmp0 = _mm_unpacklo_ps(v.val[0], v.val[1]);
+    tmp2 = _mm_unpacklo_ps(v.val[2], v.val[3]);
+    tmp1 = _mm_unpackhi_ps(v.val[0], v.val[1]);
+    tmp3 = _mm_unpackhi_ps(v.val[2], v.val[3]);
+    v.val[0] = _mm_movelh_ps(tmp0, tmp2);
+    v.val[1] = _mm_movehl_ps(tmp2, tmp0);
+    v.val[2] = _mm_movelh_ps(tmp1, tmp3);
+    v.val[3] = _mm_movehl_ps(tmp3, tmp1);
+    return v;
+#elif defined(GI_RVV_INTRINSICS)
+    return vlseg4e32_v_f32m1x4(ptr, GI_SIMD_LEN_BYTE / sizeof(float));
+#else
+    GI_FLOAT32_V4_t ret;
+    for (size_t i = 0; i < 4; i++) {
+        ret.val[i][0] = ptr[0 + i];
+        ret.val[i][1] = ptr[4 + i];
+        ret.val[i][2] = ptr[8 + i];
+        ret.val[i][3] = ptr[12 + i];
+    }
+
+    return ret;
+#endif
+}
+
 GI_FORCEINLINE
 void GiStoreZipFloat32V3(float* ptr, GI_FLOAT32_V3_t val) {
 #if defined(GI_NEON_INTRINSICS)

dnn/test/fallback/gi.cpp

@@ -4208,7 +4208,26 @@ TEST_F(FALLBACK, GiLoadUzipFloat32V3) {
         naive.push_back(s0[9 + i]);
     }
 
-    assert_eq((float*)&ret, naive);
+    assert_eq((float*)&ret, naive, SIMD_LEN * 3);
+}
+
+TEST_F(FALLBACK, GiLoadUzipFloat32V4) {
+    GI_FLOAT32_V4_t ret;
+    std::vector<float> s0{1.1f, 2.2f, 3.5f,  4.9f,  2312.1f, 345.244f, 3.59f, -12.8f,
+                          2.2f, 6.0f, 90.0f, 89.3f, 2.1f,    -3.5f,    4.9f,  -2312.1f};
+    s0.resize(SIMD_LEN * 4);
+
+    force_memset_ret((void*)&ret, GI_SIMD_LEN_BYTE * 4);
+    ret = GiLoadUzipFloat32V4(s0.data());
+    std::vector<float> naive;
+    for (size_t i = 0; i < 4; i++) {
+        naive.push_back(s0[0 + i]);
+        naive.push_back(s0[4 + i]);
+        naive.push_back(s0[8 + i]);
+        naive.push_back(s0[12 + i]);
+    }
+
+    assert_eq((float*)&ret, naive, SIMD_LEN * 4);
 }
 
 TEST_F(FALLBACK, GiStoreZipFloat32V3) {