Improve idct16x16: _256_add_sse2(x1.107)&_10_add_sse2(x1.012)

The performance gain of idct16x16_10_add_sse2 function is not
noticeable. However since both functions use the IDCT16_1D,
idct16x16_10_add_sse2 should be modified as well.
Tested with: park_joy_420_720p50.y4m

Change-Id: I02b957e36fcf997c677d15baf496533895271bff

diff --git a/vp9/common/x86/vp9_idct_intrin_sse2.c b/vp9/common/x86/vp9_idct_intrin_sse2.c
index 8fdf19a..947c0ba 100644 (file)
--- a/vp9/common/x86/vp9_idct_intrin_sse2.c
+++ b/vp9/common/x86/vp9_idct_intrin_sse2.c
@@ -650,6 +650,25 @@ static INLINE void array_transpose_8x8(__m128i *in, __m128i *res) {
   res[7] = _mm_unpackhi_epi64(tr1_6, tr1_7);
 }
 
+static INLINE void array_transpose_4X8(__m128i *in, __m128i * out) {
+  const __m128i zero = _mm_setzero_si128();
+  const __m128i tr0_0 = _mm_unpacklo_epi16(in[0], in[1]);
+  const __m128i tr0_1 = _mm_unpacklo_epi16(in[2], in[3]);
+  const __m128i tr0_4 = _mm_unpacklo_epi16(in[4], in[5]);
+  const __m128i tr0_5 = _mm_unpacklo_epi16(in[6], in[7]);
+
+  const __m128i tr1_0 = _mm_unpacklo_epi32(tr0_0, tr0_1);
+  const __m128i tr1_2 = _mm_unpackhi_epi32(tr0_0, tr0_1);
+  const __m128i tr1_4 = _mm_unpacklo_epi32(tr0_4, tr0_5);
+  const __m128i tr1_6 = _mm_unpackhi_epi32(tr0_4, tr0_5);
+
+  out[0] = _mm_unpacklo_epi64(tr1_0, tr1_4);
+  out[1] = _mm_unpackhi_epi64(tr1_0, tr1_4);
+  out[2] = _mm_unpacklo_epi64(tr1_2, tr1_6);
+  out[3] = _mm_unpackhi_epi64(tr1_2, tr1_6);
+  out[4] = out[5] = out[6] = out[7] = zero;
+}
+
 static void idct8_1d_sse2(__m128i *in) {
   const __m128i rounding = _mm_set1_epi32(DCT_CONST_ROUNDING);
   const __m128i stg1_0 = pair_set_epi16(cospi_28_64, -cospi_4_64);
@@ -1139,14 +1158,14 @@ void vp9_idct8x8_10_add_sse2(const int16_t *input, uint8_t *dest, int stride) {
 #define IDCT16_1D \
   /* Stage2 */ \
   { \
-    const __m128i lo_1_15 = _mm_unpacklo_epi16(in1, in15); \
-    const __m128i hi_1_15 = _mm_unpackhi_epi16(in1, in15); \
-    const __m128i lo_9_7 = _mm_unpacklo_epi16(in9, in7);   \
-    const __m128i hi_9_7 = _mm_unpackhi_epi16(in9, in7);   \
-    const __m128i lo_5_11 = _mm_unpacklo_epi16(in5, in11); \
-    const __m128i hi_5_11 = _mm_unpackhi_epi16(in5, in11); \
-    const __m128i lo_13_3 = _mm_unpacklo_epi16(in13, in3); \
-    const __m128i hi_13_3 = _mm_unpackhi_epi16(in13, in3); \
+    const __m128i lo_1_15 = _mm_unpacklo_epi16(in[1], in[15]); \
+    const __m128i hi_1_15 = _mm_unpackhi_epi16(in[1], in[15]); \
+    const __m128i lo_9_7 = _mm_unpacklo_epi16(in[9], in[7]);   \
+    const __m128i hi_9_7 = _mm_unpackhi_epi16(in[9], in[7]);   \
+    const __m128i lo_5_11 = _mm_unpacklo_epi16(in[5], in[11]); \
+    const __m128i hi_5_11 = _mm_unpackhi_epi16(in[5], in[11]); \
+    const __m128i lo_13_3 = _mm_unpacklo_epi16(in[13], in[3]); \
+    const __m128i hi_13_3 = _mm_unpackhi_epi16(in[13], in[3]); \
     \
     MULTIPLICATION_AND_ADD(lo_1_15, hi_1_15, lo_9_7, hi_9_7, \
                            stg2_0, stg2_1, stg2_2, stg2_3, \
@@ -1159,10 +1178,10 @@ void vp9_idct8x8_10_add_sse2(const int16_t *input, uint8_t *dest, int stride) {
     \
   /* Stage3 */ \
   { \
-    const __m128i lo_2_14 = _mm_unpacklo_epi16(in2, in14); \
-    const __m128i hi_2_14 = _mm_unpackhi_epi16(in2, in14); \
-    const __m128i lo_10_6 = _mm_unpacklo_epi16(in10, in6); \
-    const __m128i hi_10_6 = _mm_unpackhi_epi16(in10, in6); \
+    const __m128i lo_2_14 = _mm_unpacklo_epi16(in[2], in[14]); \
+    const __m128i hi_2_14 = _mm_unpackhi_epi16(in[2], in[14]); \
+    const __m128i lo_10_6 = _mm_unpacklo_epi16(in[10], in[6]); \
+    const __m128i hi_10_6 = _mm_unpackhi_epi16(in[10], in[6]); \
     \
     MULTIPLICATION_AND_ADD(lo_2_14, hi_2_14, lo_10_6, hi_10_6, \
                            stg3_0, stg3_1, stg3_2, stg3_3, \
@@ -1181,10 +1200,10 @@ void vp9_idct8x8_10_add_sse2(const int16_t *input, uint8_t *dest, int stride) {
   \
   /* Stage4 */ \
   { \
-    const __m128i lo_0_8 = _mm_unpacklo_epi16(in0, in8); \
-    const __m128i hi_0_8 = _mm_unpackhi_epi16(in0, in8); \
-    const __m128i lo_4_12 = _mm_unpacklo_epi16(in4, in12); \
-    const __m128i hi_4_12 = _mm_unpackhi_epi16(in4, in12); \
+    const __m128i lo_0_8 = _mm_unpacklo_epi16(in[0], in[8]); \
+    const __m128i hi_0_8 = _mm_unpackhi_epi16(in[0], in[8]); \
+    const __m128i lo_4_12 = _mm_unpacklo_epi16(in[4], in[12]); \
+    const __m128i hi_4_12 = _mm_unpackhi_epi16(in[4], in[12]); \
     \
     const __m128i lo_9_14 = _mm_unpacklo_epi16(stp1_9, stp1_14); \
     const __m128i hi_9_14 = _mm_unpackhi_epi16(stp1_9, stp1_14); \
@@ -1296,16 +1315,7 @@ void vp9_idct16x16_256_add_sse2(const int16_t *input, uint8_t *dest,
 
   const __m128i stg6_0 = pair_set_epi16(-cospi_16_64, cospi_16_64);
 
-  __m128i in0 = zero, in1 = zero, in2 = zero, in3 = zero, in4 = zero,
-          in5 = zero, in6 = zero, in7 = zero, in8 = zero, in9 = zero,
-          in10 = zero, in11 = zero, in12 = zero, in13 = zero,
-          in14 = zero, in15 = zero;
-  __m128i l0 = zero, l1 = zero, l2 = zero, l3 = zero, l4 = zero, l5 = zero,
-          l6 = zero, l7 = zero, l8 = zero, l9 = zero, l10 = zero, l11 = zero,
-          l12 = zero, l13 = zero, l14 = zero, l15 = zero;
-  __m128i r0 = zero, r1 = zero, r2 = zero, r3 = zero, r4 = zero, r5 = zero,
-          r6 = zero, r7 = zero, r8 = zero, r9 = zero, r10 = zero, r11 = zero,
-          r12 = zero, r13 = zero, r14 = zero, r15 = zero;
+  __m128i in[16], l[16], r[16], *curr1;
   __m128i stp1_0, stp1_1, stp1_2, stp1_3, stp1_4, stp1_5, stp1_6, stp1_7,
           stp1_8, stp1_9, stp1_10, stp1_11, stp1_12, stp1_13, stp1_14, stp1_15,
           stp1_8_0, stp1_12_0;
@@ -1314,162 +1324,132 @@ void vp9_idct16x16_256_add_sse2(const int16_t *input, uint8_t *dest,
   __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
   int i;
 
-  // We work on a 8x16 block each time, and loop 4 times for 2-D 16x16 idct.
-  for (i = 0; i < 4; i++) {
-    // 1-D idct
-    if (i < 2) {
-      if (i == 1) input += 128;
+  curr1 = l;
+  for (i = 0; i < 2; i++) {
+      // 1-D idct
 
       // Load input data.
-      in0 = _mm_load_si128((const __m128i *)input);
-      in8 = _mm_load_si128((const __m128i *)(input + 8 * 1));
-      in1 = _mm_load_si128((const __m128i *)(input + 8 * 2));
-      in9 = _mm_load_si128((const __m128i *)(input + 8 * 3));
-      in2 = _mm_load_si128((const __m128i *)(input + 8 * 4));
-      in10 = _mm_load_si128((const __m128i *)(input + 8 * 5));
-      in3 = _mm_load_si128((const __m128i *)(input + 8 * 6));
-      in11 = _mm_load_si128((const __m128i *)(input + 8 * 7));
-      in4 = _mm_load_si128((const __m128i *)(input + 8 * 8));
-      in12 = _mm_load_si128((const __m128i *)(input + 8 * 9));
-      in5 = _mm_load_si128((const __m128i *)(input + 8 * 10));
-      in13 = _mm_load_si128((const __m128i *)(input + 8 * 11));
-      in6 = _mm_load_si128((const __m128i *)(input + 8 * 12));
-      in14 = _mm_load_si128((const __m128i *)(input + 8 * 13));
-      in7 = _mm_load_si128((const __m128i *)(input + 8 * 14));
-      in15 = _mm_load_si128((const __m128i *)(input + 8 * 15));
-
-      TRANSPOSE_8X8(in0, in1, in2, in3, in4, in5, in6, in7, in0, in1, in2, in3,
-                    in4, in5, in6, in7);
-      TRANSPOSE_8X8(in8, in9, in10, in11, in12, in13, in14, in15, in8, in9,
-                    in10, in11, in12, in13, in14, in15);
-    }
+      in[0] = _mm_load_si128((const __m128i *)input);
+      in[8] = _mm_load_si128((const __m128i *)(input + 8 * 1));
+      in[1] = _mm_load_si128((const __m128i *)(input + 8 * 2));
+      in[9] = _mm_load_si128((const __m128i *)(input + 8 * 3));
+      in[2] = _mm_load_si128((const __m128i *)(input + 8 * 4));
+      in[10] = _mm_load_si128((const __m128i *)(input + 8 * 5));
+      in[3] = _mm_load_si128((const __m128i *)(input + 8 * 6));
+      in[11] = _mm_load_si128((const __m128i *)(input + 8 * 7));
+      in[4] = _mm_load_si128((const __m128i *)(input + 8 * 8));
+      in[12] = _mm_load_si128((const __m128i *)(input + 8 * 9));
+      in[5] = _mm_load_si128((const __m128i *)(input + 8 * 10));
+      in[13] = _mm_load_si128((const __m128i *)(input + 8 * 11));
+      in[6] = _mm_load_si128((const __m128i *)(input + 8 * 12));
+      in[14] = _mm_load_si128((const __m128i *)(input + 8 * 13));
+      in[7] = _mm_load_si128((const __m128i *)(input + 8 * 14));
+      in[15] = _mm_load_si128((const __m128i *)(input + 8 * 15));
 
-    if (i == 2) {
-      TRANSPOSE_8X8(l0, l1, l2, l3, l4, l5, l6, l7, in0, in1, in2, in3, in4,
-                    in5, in6, in7);
-      TRANSPOSE_8X8(r0, r1, r2, r3, r4, r5, r6, r7, in8, in9, in10, in11, in12,
-                    in13, in14, in15);
-    }
+      array_transpose_8x8(in, in);
+      array_transpose_8x8(in+8, in+8);
 
-    if (i == 3) {
-      TRANSPOSE_8X8(l8, l9, l10, l11, l12, l13, l14, l15, in0, in1, in2, in3,
-                    in4, in5, in6, in7);
-      TRANSPOSE_8X8(r8, r9, r10, r11, r12, r13, r14, r15, in8, in9, in10, in11,
-                    in12, in13, in14, in15);
-    }
+      IDCT16_1D
+
+      // Stage7
+      curr1[0] = _mm_add_epi16(stp2_0, stp1_15);
+      curr1[1] = _mm_add_epi16(stp2_1, stp1_14);
+      curr1[2] = _mm_add_epi16(stp2_2, stp2_13);
+      curr1[3] = _mm_add_epi16(stp2_3, stp2_12);
+      curr1[4] = _mm_add_epi16(stp2_4, stp2_11);
+      curr1[5] = _mm_add_epi16(stp2_5, stp2_10);
+      curr1[6] = _mm_add_epi16(stp2_6, stp1_9);
+      curr1[7] = _mm_add_epi16(stp2_7, stp1_8);
+      curr1[8] = _mm_sub_epi16(stp2_7, stp1_8);
+      curr1[9] = _mm_sub_epi16(stp2_6, stp1_9);
+      curr1[10] = _mm_sub_epi16(stp2_5, stp2_10);
+      curr1[11] = _mm_sub_epi16(stp2_4, stp2_11);
+      curr1[12] = _mm_sub_epi16(stp2_3, stp2_12);
+      curr1[13] = _mm_sub_epi16(stp2_2, stp2_13);
+      curr1[14] = _mm_sub_epi16(stp2_1, stp1_14);
+      curr1[15] = _mm_sub_epi16(stp2_0, stp1_15);
+
+      curr1 = r;
+      input += 128;
+  }
+  for (i = 0; i < 2; i++) {
+      // 1-D idct
+      array_transpose_8x8(l+i*8, in);
+      array_transpose_8x8(r+i*8, in+8);
 
-    IDCT16_1D
+      IDCT16_1D
 
-    // Stage7
-    if (i == 0) {
-      // Left 8x16
-      l0 = _mm_add_epi16(stp2_0, stp1_15);
-      l1 = _mm_add_epi16(stp2_1, stp1_14);
-      l2 = _mm_add_epi16(stp2_2, stp2_13);
-      l3 = _mm_add_epi16(stp2_3, stp2_12);
-      l4 = _mm_add_epi16(stp2_4, stp2_11);
-      l5 = _mm_add_epi16(stp2_5, stp2_10);
-      l6 = _mm_add_epi16(stp2_6, stp1_9);
-      l7 = _mm_add_epi16(stp2_7, stp1_8);
-      l8 = _mm_sub_epi16(stp2_7, stp1_8);
-      l9 = _mm_sub_epi16(stp2_6, stp1_9);
-      l10 = _mm_sub_epi16(stp2_5, stp2_10);
-      l11 = _mm_sub_epi16(stp2_4, stp2_11);
-      l12 = _mm_sub_epi16(stp2_3, stp2_12);
-      l13 = _mm_sub_epi16(stp2_2, stp2_13);
-      l14 = _mm_sub_epi16(stp2_1, stp1_14);
-      l15 = _mm_sub_epi16(stp2_0, stp1_15);
-    } else if (i == 1) {
-      // Right 8x16
-      r0 = _mm_add_epi16(stp2_0, stp1_15);
-      r1 = _mm_add_epi16(stp2_1, stp1_14);
-      r2 = _mm_add_epi16(stp2_2, stp2_13);
-      r3 = _mm_add_epi16(stp2_3, stp2_12);
-      r4 = _mm_add_epi16(stp2_4, stp2_11);
-      r5 = _mm_add_epi16(stp2_5, stp2_10);
-      r6 = _mm_add_epi16(stp2_6, stp1_9);
-      r7 = _mm_add_epi16(stp2_7, stp1_8);
-      r8 = _mm_sub_epi16(stp2_7, stp1_8);
-      r9 = _mm_sub_epi16(stp2_6, stp1_9);
-      r10 = _mm_sub_epi16(stp2_5, stp2_10);
-      r11 = _mm_sub_epi16(stp2_4, stp2_11);
-      r12 = _mm_sub_epi16(stp2_3, stp2_12);
-      r13 = _mm_sub_epi16(stp2_2, stp2_13);
-      r14 = _mm_sub_epi16(stp2_1, stp1_14);
-      r15 = _mm_sub_epi16(stp2_0, stp1_15);
-    } else {
       // 2-D
-      in0 = _mm_add_epi16(stp2_0, stp1_15);
-      in1 = _mm_add_epi16(stp2_1, stp1_14);
-      in2 = _mm_add_epi16(stp2_2, stp2_13);
-      in3 = _mm_add_epi16(stp2_3, stp2_12);
-      in4 = _mm_add_epi16(stp2_4, stp2_11);
-      in5 = _mm_add_epi16(stp2_5, stp2_10);
-      in6 = _mm_add_epi16(stp2_6, stp1_9);
-      in7 = _mm_add_epi16(stp2_7, stp1_8);
-      in8 = _mm_sub_epi16(stp2_7, stp1_8);
-      in9 = _mm_sub_epi16(stp2_6, stp1_9);
-      in10 = _mm_sub_epi16(stp2_5, stp2_10);
-      in11 = _mm_sub_epi16(stp2_4, stp2_11);
-      in12 = _mm_sub_epi16(stp2_3, stp2_12);
-      in13 = _mm_sub_epi16(stp2_2, stp2_13);
-      in14 = _mm_sub_epi16(stp2_1, stp1_14);
-      in15 = _mm_sub_epi16(stp2_0, stp1_15);
+      in[0] = _mm_add_epi16(stp2_0, stp1_15);
+      in[1] = _mm_add_epi16(stp2_1, stp1_14);
+      in[2] = _mm_add_epi16(stp2_2, stp2_13);
+      in[3] = _mm_add_epi16(stp2_3, stp2_12);
+      in[4] = _mm_add_epi16(stp2_4, stp2_11);
+      in[5] = _mm_add_epi16(stp2_5, stp2_10);
+      in[6] = _mm_add_epi16(stp2_6, stp1_9);
+      in[7] = _mm_add_epi16(stp2_7, stp1_8);
+      in[8] = _mm_sub_epi16(stp2_7, stp1_8);
+      in[9] = _mm_sub_epi16(stp2_6, stp1_9);
+      in[10] = _mm_sub_epi16(stp2_5, stp2_10);
+      in[11] = _mm_sub_epi16(stp2_4, stp2_11);
+      in[12] = _mm_sub_epi16(stp2_3, stp2_12);
+      in[13] = _mm_sub_epi16(stp2_2, stp2_13);
+      in[14] = _mm_sub_epi16(stp2_1, stp1_14);
+      in[15] = _mm_sub_epi16(stp2_0, stp1_15);
 
       // Final rounding and shift
-      in0 = _mm_adds_epi16(in0, final_rounding);
-      in1 = _mm_adds_epi16(in1, final_rounding);
-      in2 = _mm_adds_epi16(in2, final_rounding);
-      in3 = _mm_adds_epi16(in3, final_rounding);
-      in4 = _mm_adds_epi16(in4, final_rounding);
-      in5 = _mm_adds_epi16(in5, final_rounding);
-      in6 = _mm_adds_epi16(in6, final_rounding);
-      in7 = _mm_adds_epi16(in7, final_rounding);
-      in8 = _mm_adds_epi16(in8, final_rounding);
-      in9 = _mm_adds_epi16(in9, final_rounding);
-      in10 = _mm_adds_epi16(in10, final_rounding);
-      in11 = _mm_adds_epi16(in11, final_rounding);
-      in12 = _mm_adds_epi16(in12, final_rounding);
-      in13 = _mm_adds_epi16(in13, final_rounding);
-      in14 = _mm_adds_epi16(in14, final_rounding);
-      in15 = _mm_adds_epi16(in15, final_rounding);
-
-      in0 = _mm_srai_epi16(in0, 6);
-      in1 = _mm_srai_epi16(in1, 6);
-      in2 = _mm_srai_epi16(in2, 6);
-      in3 = _mm_srai_epi16(in3, 6);
-      in4 = _mm_srai_epi16(in4, 6);
-      in5 = _mm_srai_epi16(in5, 6);
-      in6 = _mm_srai_epi16(in6, 6);
-      in7 = _mm_srai_epi16(in7, 6);
-      in8 = _mm_srai_epi16(in8, 6);
-      in9 = _mm_srai_epi16(in9, 6);
-      in10 = _mm_srai_epi16(in10, 6);
-      in11 = _mm_srai_epi16(in11, 6);
-      in12 = _mm_srai_epi16(in12, 6);
-      in13 = _mm_srai_epi16(in13, 6);
-      in14 = _mm_srai_epi16(in14, 6);
-      in15 = _mm_srai_epi16(in15, 6);
-
-      RECON_AND_STORE(dest, in0);
-      RECON_AND_STORE(dest, in1);
-      RECON_AND_STORE(dest, in2);
-      RECON_AND_STORE(dest, in3);
-      RECON_AND_STORE(dest, in4);
-      RECON_AND_STORE(dest, in5);
-      RECON_AND_STORE(dest, in6);
-      RECON_AND_STORE(dest, in7);
-      RECON_AND_STORE(dest, in8);
-      RECON_AND_STORE(dest, in9);
-      RECON_AND_STORE(dest, in10);
-      RECON_AND_STORE(dest, in11);
-      RECON_AND_STORE(dest, in12);
-      RECON_AND_STORE(dest, in13);
-      RECON_AND_STORE(dest, in14);
-      RECON_AND_STORE(dest, in15);
+      in[0] = _mm_adds_epi16(in[0], final_rounding);
+      in[1] = _mm_adds_epi16(in[1], final_rounding);
+      in[2] = _mm_adds_epi16(in[2], final_rounding);
+      in[3] = _mm_adds_epi16(in[3], final_rounding);
+      in[4] = _mm_adds_epi16(in[4], final_rounding);
+      in[5] = _mm_adds_epi16(in[5], final_rounding);
+      in[6] = _mm_adds_epi16(in[6], final_rounding);
+      in[7] = _mm_adds_epi16(in[7], final_rounding);
+      in[8] = _mm_adds_epi16(in[8], final_rounding);
+      in[9] = _mm_adds_epi16(in[9], final_rounding);
+      in[10] = _mm_adds_epi16(in[10], final_rounding);
+      in[11] = _mm_adds_epi16(in[11], final_rounding);
+      in[12] = _mm_adds_epi16(in[12], final_rounding);
+      in[13] = _mm_adds_epi16(in[13], final_rounding);
+      in[14] = _mm_adds_epi16(in[14], final_rounding);
+      in[15] = _mm_adds_epi16(in[15], final_rounding);
+
+      in[0] = _mm_srai_epi16(in[0], 6);
+      in[1] = _mm_srai_epi16(in[1], 6);
+      in[2] = _mm_srai_epi16(in[2], 6);
+      in[3] = _mm_srai_epi16(in[3], 6);
+      in[4] = _mm_srai_epi16(in[4], 6);
+      in[5] = _mm_srai_epi16(in[5], 6);
+      in[6] = _mm_srai_epi16(in[6], 6);
+      in[7] = _mm_srai_epi16(in[7], 6);
+      in[8] = _mm_srai_epi16(in[8], 6);
+      in[9] = _mm_srai_epi16(in[9], 6);
+      in[10] = _mm_srai_epi16(in[10], 6);
+      in[11] = _mm_srai_epi16(in[11], 6);
+      in[12] = _mm_srai_epi16(in[12], 6);
+      in[13] = _mm_srai_epi16(in[13], 6);
+      in[14] = _mm_srai_epi16(in[14], 6);
+      in[15] = _mm_srai_epi16(in[15], 6);
+
+      RECON_AND_STORE(dest, in[0]);
+      RECON_AND_STORE(dest, in[1]);
+      RECON_AND_STORE(dest, in[2]);
+      RECON_AND_STORE(dest, in[3]);
+      RECON_AND_STORE(dest, in[4]);
+      RECON_AND_STORE(dest, in[5]);
+      RECON_AND_STORE(dest, in[6]);
+      RECON_AND_STORE(dest, in[7]);
+      RECON_AND_STORE(dest, in[8]);
+      RECON_AND_STORE(dest, in[9]);
+      RECON_AND_STORE(dest, in[10]);
+      RECON_AND_STORE(dest, in[11]);
+      RECON_AND_STORE(dest, in[12]);
+      RECON_AND_STORE(dest, in[13]);
+      RECON_AND_STORE(dest, in[14]);
+      RECON_AND_STORE(dest, in[15]);
 
       dest += 8 - (stride * 16);
-    }
   }
 }
 
@@ -2489,15 +2469,7 @@ void vp9_idct16x16_10_add_sse2(const int16_t *input, uint8_t *dest,
   const __m128i stg4_7 = pair_set_epi16(-cospi_8_64, cospi_24_64);
 
   const __m128i stg6_0 = pair_set_epi16(-cospi_16_64, cospi_16_64);
-
-  __m128i in0 = zero, in1 = zero, in2 = zero, in3 = zero, in4 = zero,
-          in5 = zero, in6 = zero, in7 = zero, in8 = zero, in9 = zero,
-          in10 = zero, in11 = zero, in12 = zero, in13 = zero,
-          in14 = zero, in15 = zero;
-  __m128i l0 = zero, l1 = zero, l2 = zero, l3 = zero, l4 = zero, l5 = zero,
-          l6 = zero, l7 = zero, l8 = zero, l9 = zero, l10 = zero, l11 = zero,
-          l12 = zero, l13 = zero, l14 = zero, l15 = zero;
-
+  __m128i in[16], l[16];
   __m128i stp1_0, stp1_1, stp1_2, stp1_3, stp1_4, stp1_5, stp1_6, stp1_7,
           stp1_8, stp1_9, stp1_10, stp1_11, stp1_12, stp1_13, stp1_14, stp1_15,
           stp1_8_0, stp1_12_0;
@@ -2505,25 +2477,26 @@ void vp9_idct16x16_10_add_sse2(const int16_t *input, uint8_t *dest,
           stp2_8, stp2_9, stp2_10, stp2_11, stp2_12, stp2_13, stp2_14, stp2_15;
   __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
   int i;
+  in[4] = in[5] = in[6] = in[7] = in[12] = in[13] = in[14] = in[15] = zero;
   // 1-D idct. Load input data.
-  in0 = _mm_load_si128((const __m128i *)input);
-  in8 = _mm_load_si128((const __m128i *)(input + 8 * 1));
-  in1 = _mm_load_si128((const __m128i *)(input + 8 * 2));
-  in9 = _mm_load_si128((const __m128i *)(input + 8 * 3));
-  in2 = _mm_load_si128((const __m128i *)(input + 8 * 4));
-  in10 = _mm_load_si128((const __m128i *)(input + 8 * 5));
-  in3 = _mm_load_si128((const __m128i *)(input + 8 * 6));
-  in11 = _mm_load_si128((const __m128i *)(input + 8 * 7));
+  in[0] = _mm_load_si128((const __m128i *)input);
+  in[8] = _mm_load_si128((const __m128i *)(input + 8 * 1));
+  in[1] = _mm_load_si128((const __m128i *)(input + 8 * 2));
+  in[9] = _mm_load_si128((const __m128i *)(input + 8 * 3));
+  in[2] = _mm_load_si128((const __m128i *)(input + 8 * 4));
+  in[10] = _mm_load_si128((const __m128i *)(input + 8 * 5));
+  in[3] = _mm_load_si128((const __m128i *)(input + 8 * 6));
+  in[11] = _mm_load_si128((const __m128i *)(input + 8 * 7));
 
-  TRANSPOSE_8X4(in0, in1, in2, in3, in0, in1, in2, in3);
-  TRANSPOSE_8X4(in8, in9, in10, in11, in8, in9, in10, in11);
+  TRANSPOSE_8X4(in[0], in[1], in[2], in[3], in[0], in[1], in[2], in[3]);
+  TRANSPOSE_8X4(in[8], in[9], in[10], in[11], in[8], in[9], in[10], in[11]);
 
   // Stage2
   {
-    const __m128i lo_1_15 = _mm_unpackhi_epi16(in0, in11);
-    const __m128i lo_9_7 = _mm_unpackhi_epi16(in8, in3);
-    const __m128i lo_5_11 = _mm_unpackhi_epi16(in2, in9);
-    const __m128i lo_13_3 = _mm_unpackhi_epi16(in10, in1);
+    const __m128i lo_1_15 = _mm_unpackhi_epi16(in[0], in[11]);
+    const __m128i lo_9_7 = _mm_unpackhi_epi16(in[8], in[3]);
+    const __m128i lo_5_11 = _mm_unpackhi_epi16(in[2], in[9]);
+    const __m128i lo_13_3 = _mm_unpackhi_epi16(in[10], in[1]);
 
     tmp0 = _mm_madd_epi16(lo_1_15, stg2_0);
     tmp2 = _mm_madd_epi16(lo_1_15, stg2_1);
@@ -2565,8 +2538,8 @@ void vp9_idct16x16_10_add_sse2(const int16_t *input, uint8_t *dest,
 
   // Stage3
   {
-    const __m128i lo_2_14 = _mm_unpacklo_epi16(in1, in11);
-    const __m128i lo_10_6 = _mm_unpacklo_epi16(in9, in3);
+    const __m128i lo_2_14 = _mm_unpacklo_epi16(in[1], in[11]);
+    const __m128i lo_10_6 = _mm_unpacklo_epi16(in[9], in[3]);
 
     tmp0 = _mm_madd_epi16(lo_2_14, stg3_0);
     tmp2 = _mm_madd_epi16(lo_2_14, stg3_1);
@@ -2601,8 +2574,8 @@ void vp9_idct16x16_10_add_sse2(const int16_t *input, uint8_t *dest,
 
   // Stage4
   {
-    const __m128i lo_0_8 = _mm_unpacklo_epi16(in0, in8);
-    const __m128i lo_4_12 = _mm_unpacklo_epi16(in2, in10);
+    const __m128i lo_0_8 = _mm_unpacklo_epi16(in[0], in[8]);
+    const __m128i lo_4_12 = _mm_unpacklo_epi16(in[2], in[10]);
     const __m128i lo_9_14 = _mm_unpacklo_epi16(stp1_9, stp1_14);
     const __m128i lo_10_13 = _mm_unpacklo_epi16(stp1_10, stp1_13);
 
@@ -2711,106 +2684,99 @@ void vp9_idct16x16_10_add_sse2(const int16_t *input, uint8_t *dest,
   }
 
   // Stage7. Left 8x16 only.
-  l0 = _mm_add_epi16(stp2_0, stp1_15);
-  l1 = _mm_add_epi16(stp2_1, stp1_14);
-  l2 = _mm_add_epi16(stp2_2, stp2_13);
-  l3 = _mm_add_epi16(stp2_3, stp2_12);
-  l4 = _mm_add_epi16(stp2_4, stp2_11);
-  l5 = _mm_add_epi16(stp2_5, stp2_10);
-  l6 = _mm_add_epi16(stp2_6, stp1_9);
-  l7 = _mm_add_epi16(stp2_7, stp1_8);
-  l8 = _mm_sub_epi16(stp2_7, stp1_8);
-  l9 = _mm_sub_epi16(stp2_6, stp1_9);
-  l10 = _mm_sub_epi16(stp2_5, stp2_10);
-  l11 = _mm_sub_epi16(stp2_4, stp2_11);
-  l12 = _mm_sub_epi16(stp2_3, stp2_12);
-  l13 = _mm_sub_epi16(stp2_2, stp2_13);
-  l14 = _mm_sub_epi16(stp2_1, stp1_14);
-  l15 = _mm_sub_epi16(stp2_0, stp1_15);
+  l[0] = _mm_add_epi16(stp2_0, stp1_15);
+  l[1] = _mm_add_epi16(stp2_1, stp1_14);
+  l[2] = _mm_add_epi16(stp2_2, stp2_13);
+  l[3] = _mm_add_epi16(stp2_3, stp2_12);
+  l[4] = _mm_add_epi16(stp2_4, stp2_11);
+  l[5] = _mm_add_epi16(stp2_5, stp2_10);
+  l[6] = _mm_add_epi16(stp2_6, stp1_9);
+  l[7] = _mm_add_epi16(stp2_7, stp1_8);
+  l[8] = _mm_sub_epi16(stp2_7, stp1_8);
+  l[9] = _mm_sub_epi16(stp2_6, stp1_9);
+  l[10] = _mm_sub_epi16(stp2_5, stp2_10);
+  l[11] = _mm_sub_epi16(stp2_4, stp2_11);
+  l[12] = _mm_sub_epi16(stp2_3, stp2_12);
+  l[13] = _mm_sub_epi16(stp2_2, stp2_13);
+  l[14] = _mm_sub_epi16(stp2_1, stp1_14);
+  l[15] = _mm_sub_epi16(stp2_0, stp1_15);
 
   // 2-D idct. We do 2 8x16 blocks.
   for (i = 0; i < 2; i++) {
-    if (i == 0)
-      TRANSPOSE_4X8(l0, l1, l2, l3, l4, l5, l6, l7, in0, in1, in2, in3, in4,
-                    in5, in6, in7);
-
-    if (i == 1)
-      TRANSPOSE_4X8(l8, l9, l10, l11, l12, l13, l14, l15, in0, in1, in2, in3,
-                    in4, in5, in6, in7);
-
-    in8 = in9 = in10 = in11 = in12 = in13 = in14 = in15 = zero;
+    array_transpose_4X8(l + 8*i, in);
+    in[8] = in[9] = in[10] = in[11] = in[12] = in[13] = in[14] = in[15] = zero;
 
     IDCT16_1D
 
     // Stage7
-    in0 = _mm_add_epi16(stp2_0, stp1_15);
-    in1 = _mm_add_epi16(stp2_1, stp1_14);
-    in2 = _mm_add_epi16(stp2_2, stp2_13);
-    in3 = _mm_add_epi16(stp2_3, stp2_12);
-    in4 = _mm_add_epi16(stp2_4, stp2_11);
-    in5 = _mm_add_epi16(stp2_5, stp2_10);
-    in6 = _mm_add_epi16(stp2_6, stp1_9);
-    in7 = _mm_add_epi16(stp2_7, stp1_8);
-    in8 = _mm_sub_epi16(stp2_7, stp1_8);
-    in9 = _mm_sub_epi16(stp2_6, stp1_9);
-    in10 = _mm_sub_epi16(stp2_5, stp2_10);
-    in11 = _mm_sub_epi16(stp2_4, stp2_11);
-    in12 = _mm_sub_epi16(stp2_3, stp2_12);
-    in13 = _mm_sub_epi16(stp2_2, stp2_13);
-    in14 = _mm_sub_epi16(stp2_1, stp1_14);
-    in15 = _mm_sub_epi16(stp2_0, stp1_15);
+    in[0] = _mm_add_epi16(stp2_0, stp1_15);
+    in[1] = _mm_add_epi16(stp2_1, stp1_14);
+    in[2] = _mm_add_epi16(stp2_2, stp2_13);
+    in[3] = _mm_add_epi16(stp2_3, stp2_12);
+    in[4] = _mm_add_epi16(stp2_4, stp2_11);
+    in[5] = _mm_add_epi16(stp2_5, stp2_10);
+    in[6] = _mm_add_epi16(stp2_6, stp1_9);
+    in[7] = _mm_add_epi16(stp2_7, stp1_8);
+    in[8] = _mm_sub_epi16(stp2_7, stp1_8);
+    in[9] = _mm_sub_epi16(stp2_6, stp1_9);
+    in[10] = _mm_sub_epi16(stp2_5, stp2_10);
+    in[11] = _mm_sub_epi16(stp2_4, stp2_11);
+    in[12] = _mm_sub_epi16(stp2_3, stp2_12);
+    in[13] = _mm_sub_epi16(stp2_2, stp2_13);
+    in[14] = _mm_sub_epi16(stp2_1, stp1_14);
+    in[15] = _mm_sub_epi16(stp2_0, stp1_15);
 
     // Final rounding and shift
-    in0 = _mm_adds_epi16(in0, final_rounding);
-    in1 = _mm_adds_epi16(in1, final_rounding);
-    in2 = _mm_adds_epi16(in2, final_rounding);
-    in3 = _mm_adds_epi16(in3, final_rounding);
-    in4 = _mm_adds_epi16(in4, final_rounding);
-    in5 = _mm_adds_epi16(in5, final_rounding);
-    in6 = _mm_adds_epi16(in6, final_rounding);
-    in7 = _mm_adds_epi16(in7, final_rounding);
-    in8 = _mm_adds_epi16(in8, final_rounding);
-    in9 = _mm_adds_epi16(in9, final_rounding);
-    in10 = _mm_adds_epi16(in10, final_rounding);
-    in11 = _mm_adds_epi16(in11, final_rounding);
-    in12 = _mm_adds_epi16(in12, final_rounding);
-    in13 = _mm_adds_epi16(in13, final_rounding);
-    in14 = _mm_adds_epi16(in14, final_rounding);
-    in15 = _mm_adds_epi16(in15, final_rounding);
-
-    in0 = _mm_srai_epi16(in0, 6);
-    in1 = _mm_srai_epi16(in1, 6);
-    in2 = _mm_srai_epi16(in2, 6);
-    in3 = _mm_srai_epi16(in3, 6);
-    in4 = _mm_srai_epi16(in4, 6);
-    in5 = _mm_srai_epi16(in5, 6);
-    in6 = _mm_srai_epi16(in6, 6);
-    in7 = _mm_srai_epi16(in7, 6);
-    in8 = _mm_srai_epi16(in8, 6);
-    in9 = _mm_srai_epi16(in9, 6);
-    in10 = _mm_srai_epi16(in10, 6);
-    in11 = _mm_srai_epi16(in11, 6);
-    in12 = _mm_srai_epi16(in12, 6);
-    in13 = _mm_srai_epi16(in13, 6);
-    in14 = _mm_srai_epi16(in14, 6);
-    in15 = _mm_srai_epi16(in15, 6);
-
-    RECON_AND_STORE(dest, in0);
-    RECON_AND_STORE(dest, in1);
-    RECON_AND_STORE(dest, in2);
-    RECON_AND_STORE(dest, in3);
-    RECON_AND_STORE(dest, in4);
-    RECON_AND_STORE(dest, in5);
-    RECON_AND_STORE(dest, in6);
-    RECON_AND_STORE(dest, in7);
-    RECON_AND_STORE(dest, in8);
-    RECON_AND_STORE(dest, in9);
-    RECON_AND_STORE(dest, in10);
-    RECON_AND_STORE(dest, in11);
-    RECON_AND_STORE(dest, in12);
-    RECON_AND_STORE(dest, in13);
-    RECON_AND_STORE(dest, in14);
-    RECON_AND_STORE(dest, in15);
+    in[0] = _mm_adds_epi16(in[0], final_rounding);
+    in[1] = _mm_adds_epi16(in[1], final_rounding);
+    in[2] = _mm_adds_epi16(in[2], final_rounding);
+    in[3] = _mm_adds_epi16(in[3], final_rounding);
+    in[4] = _mm_adds_epi16(in[4], final_rounding);
+    in[5] = _mm_adds_epi16(in[5], final_rounding);
+    in[6] = _mm_adds_epi16(in[6], final_rounding);
+    in[7] = _mm_adds_epi16(in[7], final_rounding);
+    in[8] = _mm_adds_epi16(in[8], final_rounding);
+    in[9] = _mm_adds_epi16(in[9], final_rounding);
+    in[10] = _mm_adds_epi16(in[10], final_rounding);
+    in[11] = _mm_adds_epi16(in[11], final_rounding);
+    in[12] = _mm_adds_epi16(in[12], final_rounding);
+    in[13] = _mm_adds_epi16(in[13], final_rounding);
+    in[14] = _mm_adds_epi16(in[14], final_rounding);
+    in[15] = _mm_adds_epi16(in[15], final_rounding);
+
+    in[0] = _mm_srai_epi16(in[0], 6);
+    in[1] = _mm_srai_epi16(in[1], 6);
+    in[2] = _mm_srai_epi16(in[2], 6);
+    in[3] = _mm_srai_epi16(in[3], 6);
+    in[4] = _mm_srai_epi16(in[4], 6);
+    in[5] = _mm_srai_epi16(in[5], 6);
+    in[6] = _mm_srai_epi16(in[6], 6);
+    in[7] = _mm_srai_epi16(in[7], 6);
+    in[8] = _mm_srai_epi16(in[8], 6);
+    in[9] = _mm_srai_epi16(in[9], 6);
+    in[10] = _mm_srai_epi16(in[10], 6);
+    in[11] = _mm_srai_epi16(in[11], 6);
+    in[12] = _mm_srai_epi16(in[12], 6);
+    in[13] = _mm_srai_epi16(in[13], 6);
+    in[14] = _mm_srai_epi16(in[14], 6);
+    in[15] = _mm_srai_epi16(in[15], 6);
+
+    RECON_AND_STORE(dest, in[0]);
+    RECON_AND_STORE(dest, in[1]);
+    RECON_AND_STORE(dest, in[2]);
+    RECON_AND_STORE(dest, in[3]);
+    RECON_AND_STORE(dest, in[4]);
+    RECON_AND_STORE(dest, in[5]);
+    RECON_AND_STORE(dest, in[6]);
+    RECON_AND_STORE(dest, in[7]);
+    RECON_AND_STORE(dest, in[8]);
+    RECON_AND_STORE(dest, in[9]);
+    RECON_AND_STORE(dest, in[10]);
+    RECON_AND_STORE(dest, in[11]);
+    RECON_AND_STORE(dest, in[12]);
+    RECON_AND_STORE(dest, in[13]);
+    RECON_AND_STORE(dest, in[14]);
+    RECON_AND_STORE(dest, in[15]);
 
     dest += 8 - (stride * 16);
   }