diff --git a/aom_dsp/x86/avg_intrin_sse2.c b/aom_dsp/x86/avg_intrin_sse2.c
index 67ea85b..a52abd0 100644
--- a/aom_dsp/x86/avg_intrin_sse2.c
+++ b/aom_dsp/x86/avg_intrin_sse2.c
@@ -460,17 +460,21 @@
 int aom_satd_sse2(const tran_low_t *coeff, int length) {
   int i;
   const __m128i zero = _mm_setzero_si128();
+  const __m128i one = _mm_set1_epi16(1);
   __m128i accum = zero;
 
-  for (i = 0; i < length; i += 8) {
-    const __m128i src_line = load_tran_low(coeff);
-    const __m128i inv = _mm_sub_epi16(zero, src_line);
-    const __m128i abs = _mm_max_epi16(src_line, inv);  // abs(src_line)
-    const __m128i abs_lo = _mm_unpacklo_epi16(abs, zero);
-    const __m128i abs_hi = _mm_unpackhi_epi16(abs, zero);
-    const __m128i sum = _mm_add_epi32(abs_lo, abs_hi);
-    accum = _mm_add_epi32(accum, sum);
-    coeff += 8;
+  for (i = 0; i < length; i += 16) {
+    const __m128i src_line0 = load_tran_low(coeff);
+    const __m128i src_line1 = load_tran_low(coeff + 8);
+    const __m128i inv0 = _mm_sub_epi16(zero, src_line0);
+    const __m128i inv1 = _mm_sub_epi16(zero, src_line1);
+    const __m128i abs0 = _mm_max_epi16(src_line0, inv0);  // abs(src_line)
+    const __m128i abs1 = _mm_max_epi16(src_line1, inv1);  // abs(src_line)
+    const __m128i sum0 = _mm_madd_epi16(abs0, one);
+    const __m128i sum1 = _mm_madd_epi16(abs1, one);
+    accum = _mm_add_epi32(accum, sum0);
+    accum = _mm_add_epi32(accum, sum1);
+    coeff += 16;
   }
 
   {  // cascading summation of accum