vpxdsp: [x86] add highbd_h_predictor functions

C vs SSE2 speed gains:
_4x4 : ~8.12x
_8x8 : ~9.71x
_16x16 : ~8.21x
_32x32 : ~5.0x

BUG=webm:1422

Change-Id: I5e8a1ed4db7b8dc539b3e2a728b0b34d8b4b1993

1 files changed, 159 insertions, 0 deletions

diff --git a/vpx_dsp/x86/highbd_intrapred_intrin_sse2.c b/vpx_dsp/x86/highbd_intrapred_intrin_sse2.c
new file mode 100644
index 000000000..059f900ff
--- /dev/null
+++ b/vpx_dsp/x86/highbd_intrapred_intrin_sse2.c
@@ -0,0 +1,159 @@
+/*
+ *  Copyright (c) 2017 The WebM project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include <emmintrin.h>  // SSE2
+
+#include "./vpx_config.h"
+#include "./vpx_dsp_rtcd.h"
+#include "vpx/vpx_integer.h"
+
+// -----------------------------------------------------------------------------
+
+void vpx_highbd_h_predictor_4x4_sse2(uint16_t *dst, ptrdiff_t stride,
+                                     const uint16_t *above,
+                                     const uint16_t *left, int bd) {
+  const __m128i left_u16 = _mm_loadl_epi64((const __m128i *)left);
+  const __m128i row0 = _mm_shufflelo_epi16(left_u16, 0x0);
+  const __m128i row1 = _mm_shufflelo_epi16(left_u16, 0x55);
+  const __m128i row2 = _mm_shufflelo_epi16(left_u16, 0xaa);
+  const __m128i row3 = _mm_shufflelo_epi16(left_u16, 0xff);
+  (void)above;
+  (void)bd;
+  _mm_storel_epi64((__m128i *)dst, row0);
+  dst += stride;
+  _mm_storel_epi64((__m128i *)dst, row1);
+  dst += stride;
+  _mm_storel_epi64((__m128i *)dst, row2);
+  dst += stride;
+  _mm_storel_epi64((__m128i *)dst, row3);
+}
+
+void vpx_highbd_h_predictor_8x8_sse2(uint16_t *dst, ptrdiff_t stride,
+                                     const uint16_t *above,
+                                     const uint16_t *left, int bd) {
+  const __m128i left_u16 = _mm_load_si128((const __m128i *)left);
+  const __m128i row0 = _mm_shufflelo_epi16(left_u16, 0x0);
+  const __m128i row1 = _mm_shufflelo_epi16(left_u16, 0x55);
+  const __m128i row2 = _mm_shufflelo_epi16(left_u16, 0xaa);
+  const __m128i row3 = _mm_shufflelo_epi16(left_u16, 0xff);
+  const __m128i row4 = _mm_shufflehi_epi16(left_u16, 0x0);
+  const __m128i row5 = _mm_shufflehi_epi16(left_u16, 0x55);
+  const __m128i row6 = _mm_shufflehi_epi16(left_u16, 0xaa);
+  const __m128i row7 = _mm_shufflehi_epi16(left_u16, 0xff);
+  (void)above;
+  (void)bd;
+  _mm_store_si128((__m128i *)dst, _mm_unpacklo_epi64(row0, row0));
+  dst += stride;
+  _mm_store_si128((__m128i *)dst, _mm_unpacklo_epi64(row1, row1));
+  dst += stride;
+  _mm_store_si128((__m128i *)dst, _mm_unpacklo_epi64(row2, row2));
+  dst += stride;
+  _mm_store_si128((__m128i *)dst, _mm_unpacklo_epi64(row3, row3));
+  dst += stride;
+  _mm_store_si128((__m128i *)dst, _mm_unpackhi_epi64(row4, row4));
+  dst += stride;
+  _mm_store_si128((__m128i *)dst, _mm_unpackhi_epi64(row5, row5));
+  dst += stride;
+  _mm_store_si128((__m128i *)dst, _mm_unpackhi_epi64(row6, row6));
+  dst += stride;
+  _mm_store_si128((__m128i *)dst, _mm_unpackhi_epi64(row7, row7));
+}
+
+static INLINE void h_store_16_unpacklo(uint16_t **dst, const ptrdiff_t stride,
+                                       const __m128i *row) {
+  const __m128i val = _mm_unpacklo_epi64(*row, *row);
+  _mm_store_si128((__m128i *)*dst, val);
+  _mm_store_si128((__m128i *)(*dst + 8), val);
+  *dst += stride;
+}
+
+static INLINE void h_store_16_unpackhi(uint16_t **dst, const ptrdiff_t stride,
+                                       const __m128i *row) {
+  const __m128i val = _mm_unpackhi_epi64(*row, *row);
+  _mm_store_si128((__m128i *)(*dst), val);
+  _mm_store_si128((__m128i *)(*dst + 8), val);
+  *dst += stride;
+}
+
+void vpx_highbd_h_predictor_16x16_sse2(uint16_t *dst, ptrdiff_t stride,
+                                       const uint16_t *above,
+                                       const uint16_t *left, int bd) {
+  int i;
+  (void)above;
+  (void)bd;
+
+  for (i = 0; i < 2; i++, left += 8) {
+    const __m128i left_u16 = _mm_load_si128((const __m128i *)left);
+    const __m128i row0 = _mm_shufflelo_epi16(left_u16, 0x0);
+    const __m128i row1 = _mm_shufflelo_epi16(left_u16, 0x55);
+    const __m128i row2 = _mm_shufflelo_epi16(left_u16, 0xaa);
+    const __m128i row3 = _mm_shufflelo_epi16(left_u16, 0xff);
+    const __m128i row4 = _mm_shufflehi_epi16(left_u16, 0x0);
+    const __m128i row5 = _mm_shufflehi_epi16(left_u16, 0x55);
+    const __m128i row6 = _mm_shufflehi_epi16(left_u16, 0xaa);
+    const __m128i row7 = _mm_shufflehi_epi16(left_u16, 0xff);
+    h_store_16_unpacklo(&dst, stride, &row0);
+    h_store_16_unpacklo(&dst, stride, &row1);
+    h_store_16_unpacklo(&dst, stride, &row2);
+    h_store_16_unpacklo(&dst, stride, &row3);
+    h_store_16_unpackhi(&dst, stride, &row4);
+    h_store_16_unpackhi(&dst, stride, &row5);
+    h_store_16_unpackhi(&dst, stride, &row6);
+    h_store_16_unpackhi(&dst, stride, &row7);
+  }
+}
+
+static INLINE void h_store_32_unpacklo(uint16_t **dst, const ptrdiff_t stride,
+                                       const __m128i *row) {
+  const __m128i val = _mm_unpacklo_epi64(*row, *row);
+  _mm_store_si128((__m128i *)(*dst), val);
+  _mm_store_si128((__m128i *)(*dst + 8), val);
+  _mm_store_si128((__m128i *)(*dst + 16), val);
+  _mm_store_si128((__m128i *)(*dst + 24), val);
+  *dst += stride;
+}
+
+static INLINE void h_store_32_unpackhi(uint16_t **dst, const ptrdiff_t stride,
+                                       const __m128i *row) {
+  const __m128i val = _mm_unpackhi_epi64(*row, *row);
+  _mm_store_si128((__m128i *)(*dst), val);
+  _mm_store_si128((__m128i *)(*dst + 8), val);
+  _mm_store_si128((__m128i *)(*dst + 16), val);
+  _mm_store_si128((__m128i *)(*dst + 24), val);
+  *dst += stride;
+}
+
+void vpx_highbd_h_predictor_32x32_sse2(uint16_t *dst, ptrdiff_t stride,
+                                       const uint16_t *above,
+                                       const uint16_t *left, int bd) {
+  int i;
+  (void)above;
+  (void)bd;
+
+  for (i = 0; i < 4; i++, left += 8) {
+    const __m128i left_u16 = _mm_load_si128((const __m128i *)left);
+    const __m128i row0 = _mm_shufflelo_epi16(left_u16, 0x0);
+    const __m128i row1 = _mm_shufflelo_epi16(left_u16, 0x55);
+    const __m128i row2 = _mm_shufflelo_epi16(left_u16, 0xaa);
+    const __m128i row3 = _mm_shufflelo_epi16(left_u16, 0xff);
+    const __m128i row4 = _mm_shufflehi_epi16(left_u16, 0x0);
+    const __m128i row5 = _mm_shufflehi_epi16(left_u16, 0x55);
+    const __m128i row6 = _mm_shufflehi_epi16(left_u16, 0xaa);
+    const __m128i row7 = _mm_shufflehi_epi16(left_u16, 0xff);
+    h_store_32_unpacklo(&dst, stride, &row0);
+    h_store_32_unpacklo(&dst, stride, &row1);
+    h_store_32_unpacklo(&dst, stride, &row2);
+    h_store_32_unpacklo(&dst, stride, &row3);
+    h_store_32_unpackhi(&dst, stride, &row4);
+    h_store_32_unpackhi(&dst, stride, &row5);
+    h_store_32_unpackhi(&dst, stride, &row6);
+    h_store_32_unpackhi(&dst, stride, &row7);
+  }
+}