Add vpx_dsp/x86/mem_sse2.h

Add some load and store sse2 inline functions.

Change-Id: Ib1e0650b5a3d8e2b3736ab7c7642d6e384354222

vpx_dsp/vpx_dsp.mk

@@ -386,6 +386,7 @@ DSP_SRCS-$(HAVE_VSX)  += ppc/transpose_vsx.h
 DSP_SRCS-$(HAVE_VSX)  += ppc/bitdepth_conversion_vsx.h
 
 # X86 utilities
+DSP_SRCS-$(HAVE_SSE2) += x86/mem_sse2.h
 DSP_SRCS-$(HAVE_SSE2) += x86/transpose_sse2.h
 
 DSP_SRCS-no += $(DSP_SRCS_REMOVE-yes)

vpx_dsp/x86/mem_sse2.h

+/*
+ *  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.
+ */
+
+#ifndef VPX_DSP_X86_MEM_SSE2_H_
+#define VPX_DSP_X86_MEM_SSE2_H_
+
+#include <emmintrin.h>  // SSE2
+
+#include "./vpx_config.h"
+
+static INLINE void load_8bit_4x4(const uint8_t *const s, const ptrdiff_t stride,
+                                 __m128i *const d) {
+  d[0] = _mm_cvtsi32_si128(*(const int *)(s + 0 * stride));
+  d[1] = _mm_cvtsi32_si128(*(const int *)(s + 1 * stride));
+  d[2] = _mm_cvtsi32_si128(*(const int *)(s + 2 * stride));
+  d[3] = _mm_cvtsi32_si128(*(const int *)(s + 3 * stride));
+}
+
+static INLINE void load_8bit_4x8(const uint8_t *const s, const ptrdiff_t stride,
+                                 __m128i *const d) {
+  load_8bit_4x4(s + 0 * stride, stride, &d[0]);
+  load_8bit_4x4(s + 4 * stride, stride, &d[4]);
+}
+
+static INLINE void load_8bit_8x4(const uint8_t *const s, const ptrdiff_t stride,
+                                 __m128i *const d) {
+  d[0] = _mm_loadl_epi64((const __m128i *)(s + 0 * stride));
+  d[1] = _mm_loadl_epi64((const __m128i *)(s + 1 * stride));
+  d[2] = _mm_loadl_epi64((const __m128i *)(s + 2 * stride));
+  d[3] = _mm_loadl_epi64((const __m128i *)(s + 3 * stride));
+}
+
+static INLINE void load_8bit_8x8(const uint8_t *const s, const ptrdiff_t stride,
+                                 __m128i *const d) {
+  load_8bit_8x4(s + 0 * stride, stride, &d[0]);
+  load_8bit_8x4(s + 4 * stride, stride, &d[4]);
+}
+
+static INLINE void load_8bit_16x8(const uint8_t *const s,
+                                  const ptrdiff_t stride, __m128i *const d) {
+  d[0] = _mm_load_si128((const __m128i *)(s + 0 * stride));
+  d[1] = _mm_load_si128((const __m128i *)(s + 1 * stride));
+  d[2] = _mm_load_si128((const __m128i *)(s + 2 * stride));
+  d[3] = _mm_load_si128((const __m128i *)(s + 3 * stride));
+  d[4] = _mm_load_si128((const __m128i *)(s + 4 * stride));
+  d[5] = _mm_load_si128((const __m128i *)(s + 5 * stride));
+  d[6] = _mm_load_si128((const __m128i *)(s + 6 * stride));
+  d[7] = _mm_load_si128((const __m128i *)(s + 7 * stride));
+}
+
+static INLINE void loadu_8bit_16x8(const uint8_t *const s,
+                                   const ptrdiff_t stride, __m128i *const d) {
+  d[0] = _mm_loadu_si128((const __m128i *)(s + 0 * stride));
+  d[1] = _mm_loadu_si128((const __m128i *)(s + 1 * stride));
+  d[2] = _mm_loadu_si128((const __m128i *)(s + 2 * stride));
+  d[3] = _mm_loadu_si128((const __m128i *)(s + 3 * stride));
+  d[4] = _mm_loadu_si128((const __m128i *)(s + 4 * stride));
+  d[5] = _mm_loadu_si128((const __m128i *)(s + 5 * stride));
+  d[6] = _mm_loadu_si128((const __m128i *)(s + 6 * stride));
+  d[7] = _mm_loadu_si128((const __m128i *)(s + 7 * stride));
+}
+
+static INLINE void store_8bit_4x4(const __m128i *const s, uint8_t *const d,
+                                  const ptrdiff_t stride) {
+  *(int *)(d + 0 * stride) = _mm_cvtsi128_si32(s[0]);
+  *(int *)(d + 1 * stride) = _mm_cvtsi128_si32(s[1]);
+  *(int *)(d + 2 * stride) = _mm_cvtsi128_si32(s[2]);
+  *(int *)(d + 3 * stride) = _mm_cvtsi128_si32(s[3]);
+}
+
+static INLINE void store_8bit_8x8(const __m128i *const s, uint8_t *const d,
+                                  const ptrdiff_t stride) {
+  _mm_storel_epi64((__m128i *)(d + 0 * stride), s[0]);
+  _mm_storel_epi64((__m128i *)(d + 1 * stride), s[1]);
+  _mm_storel_epi64((__m128i *)(d + 2 * stride), s[2]);
+  _mm_storel_epi64((__m128i *)(d + 3 * stride), s[3]);
+  _mm_storel_epi64((__m128i *)(d + 4 * stride), s[4]);
+  _mm_storel_epi64((__m128i *)(d + 5 * stride), s[5]);
+  _mm_storel_epi64((__m128i *)(d + 6 * stride), s[6]);
+  _mm_storel_epi64((__m128i *)(d + 7 * stride), s[7]);
+}
+
+#endif  // VPX_DSP_X86_MEM_SSE2_H_