revert: "Optimize 32 and wider avx2 block subtract (Kyle Siefring) - …

…Tests can't be added"

This reverts commit f14607b.

Source/Lib/ASM_AVX2/convolve_avx2.c

@@ -1499,17 +1499,16 @@ uint64_t svt_av1_wedge_sse_from_residuals_avx2(const int16_t *r1, const int16_t
 }
 
 static INLINE void subtract32_avx2(int16_t *diff_ptr, const uint8_t *src_ptr, const uint8_t *pred_ptr) {
-  __m256i s = _mm256_lddqu_si256((__m256i *)(src_ptr));
-  __m256i p = _mm256_lddqu_si256((__m256i *)(pred_ptr));
-  __m256i set_one_minusone = _mm256_set1_epi32((int)0xff01ff01);
-  __m256i diff0 = _mm256_unpacklo_epi8(s, p);
-  __m256i diff1 = _mm256_unpackhi_epi8(s, p);
-  diff0 = _mm256_maddubs_epi16(diff0, set_one_minusone);
-  diff1 = _mm256_maddubs_epi16(diff1, set_one_minusone);
-  _mm256_store_si256((__m256i *)(diff_ptr),
-                     _mm256_permute2x128_si256(diff0, diff1, 0x20));
-  _mm256_store_si256((__m256i *)(diff_ptr + 16),
-                     _mm256_permute2x128_si256(diff0, diff1, 0x31));
+    __m256i       s   = _mm256_lddqu_si256((__m256i *)(src_ptr));
+    __m256i       p   = _mm256_lddqu_si256((__m256i *)(pred_ptr));
+    __m256i       s_0 = _mm256_cvtepu8_epi16(_mm256_castsi256_si128(s));
+    __m256i       s_1 = _mm256_cvtepu8_epi16(_mm256_extracti128_si256(s, 1));
+    __m256i       p_0 = _mm256_cvtepu8_epi16(_mm256_castsi256_si128(p));
+    __m256i       p_1 = _mm256_cvtepu8_epi16(_mm256_extracti128_si256(p, 1));
+    const __m256i d_0 = _mm256_sub_epi16(s_0, p_0);
+    const __m256i d_1 = _mm256_sub_epi16(s_1, p_1);
+    _mm256_storeu_si256((__m256i *)(diff_ptr), d_0);
+    _mm256_storeu_si256((__m256i *)(diff_ptr + 16), d_1);
 }
 
 static INLINE void aom_subtract_block_16xn_avx2(int rows, int16_t *diff_ptr, ptrdiff_t diff_stride,