From f92d0bbbd1e7eb2bd77e1ebb7edc33a516bc29f6 Mon Sep 17 00:00:00 2001
From: "finnur@chromium.org"
 <finnur@chromium.org@0039d316-1c4b-4281-b951-d872f2087c98>
Date: Fri, 11 Feb 2011 11:58:29 +0000
Subject: Revert 74583 - Revert 74571 - Use SSE2 block differ for chromoting

(Quick test to see if it is the cause of crashes in media_unittests)
(Test showed this CL is not to blame)

We have the SSE2 lying around in the tree just never being used. This will
allow us to use it.

BUG=None
TEST=Chromoting to a host machine and the diff will work correctly

Review URL: http://codereview.chromium.org/6469022

TBR=hclam@chromium.org
Review URL: http://codereview.chromium.org/6488023

TBR=finnur@chromium.org

git-svn-id: svn://svn.chromium.org/chrome/trunk/src@74588 0039d316-1c4b-4281-b951-d872f2087c98
---
 remoting/host/differ_block_sse2.cc | 111 +++++++++++++++++++++++++++++++++++++
 1 file changed, 111 insertions(+)
 create mode 100644 remoting/host/differ_block_sse2.cc

(limited to 'remoting/host/differ_block_sse2.cc')

diff --git a/remoting/host/differ_block_sse2.cc b/remoting/host/differ_block_sse2.cc
new file mode 100644
index 0000000..c0cc3b6
--- /dev/null
+++ b/remoting/host/differ_block_sse2.cc
@@ -0,0 +1,111 @@
+// Copyright (c) 2011 The Chromium Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#if defined(_MSC_VER)
+#include <intrin.h>
+#else
+#include <mmintrin.h>
+#include <emmintrin.h>
+#endif
+
+#include "remoting/host/differ_block.h"
+#include "remoting/host/differ_block_internal.h"
+
+namespace remoting {
+
+extern int BlockDifference_SSE2_W16(const uint8* image1, const uint8* image2,
+                                    int stride) {
+  __m128i acc = _mm_setzero_si128();
+  __m128i v0;
+  __m128i v1;
+  __m128i sad;
+  for (int y = 0; y < kBlockHeight; ++y) {
+    const __m128i* i1 = reinterpret_cast<const __m128i*>(image1);
+    const __m128i* i2 = reinterpret_cast<const __m128i*>(image2);
+    v0 = _mm_loadu_si128(i1);
+    v1 = _mm_loadu_si128(i2);
+    sad = _mm_sad_epu8(v0, v1);
+    acc = _mm_adds_epu16(acc, sad);
+    v0 = _mm_loadu_si128(i1 + 1);
+    v1 = _mm_loadu_si128(i2 + 1);
+    sad = _mm_sad_epu8(v0, v1);
+    acc = _mm_adds_epu16(acc, sad);
+    v0 = _mm_loadu_si128(i1 + 2);
+    v1 = _mm_loadu_si128(i2 + 2);
+    sad = _mm_sad_epu8(v0, v1);
+    acc = _mm_adds_epu16(acc, sad);
+    v0 = _mm_loadu_si128(i1 + 3);
+    v1 = _mm_loadu_si128(i2 + 3);
+    sad = _mm_sad_epu8(v0, v1);
+    acc = _mm_adds_epu16(acc, sad);
+
+    // This essential means sad = acc >> 64. We only care about the lower 16
+    // bits.
+    sad = _mm_shuffle_epi32(acc, 0xEE);
+    sad = _mm_adds_epu16(sad, acc);
+    int diff = _mm_cvtsi128_si32(sad);
+    if (diff)
+      return 1;
+    image1 += stride;
+    image2 += stride;
+  }
+  return 0;
+}
+
+extern int BlockDifference_SSE2_W32(const uint8* image1, const uint8* image2,
+                                    int stride) {
+  __m128i acc = _mm_setzero_si128();
+  __m128i v0;
+  __m128i v1;
+  __m128i sad;
+  for (int y = 0; y < kBlockHeight; ++y) {
+    const __m128i* i1 = reinterpret_cast<const __m128i*>(image1);
+    const __m128i* i2 = reinterpret_cast<const __m128i*>(image2);
+    v0 = _mm_loadu_si128(i1);
+    v1 = _mm_loadu_si128(i2);
+    sad = _mm_sad_epu8(v0, v1);
+    acc = _mm_adds_epu16(acc, sad);
+    v0 = _mm_loadu_si128(i1 + 1);
+    v1 = _mm_loadu_si128(i2 + 1);
+    sad = _mm_sad_epu8(v0, v1);
+    acc = _mm_adds_epu16(acc, sad);
+    v0 = _mm_loadu_si128(i1 + 2);
+    v1 = _mm_loadu_si128(i2 + 2);
+    sad = _mm_sad_epu8(v0, v1);
+    acc = _mm_adds_epu16(acc, sad);
+    v0 = _mm_loadu_si128(i1 + 3);
+    v1 = _mm_loadu_si128(i2 + 3);
+    sad = _mm_sad_epu8(v0, v1);
+    acc = _mm_adds_epu16(acc, sad);
+    v0 = _mm_loadu_si128(i1 + 4);
+    v1 = _mm_loadu_si128(i2 + 4);
+    sad = _mm_sad_epu8(v0, v1);
+    acc = _mm_adds_epu16(acc, sad);
+    v0 = _mm_loadu_si128(i1 + 5);
+    v1 = _mm_loadu_si128(i2 + 5);
+    sad = _mm_sad_epu8(v0, v1);
+    acc = _mm_adds_epu16(acc, sad);
+    v0 = _mm_loadu_si128(i1 + 6);
+    v1 = _mm_loadu_si128(i2 + 6);
+    sad = _mm_sad_epu8(v0, v1);
+    acc = _mm_adds_epu16(acc, sad);
+    v0 = _mm_loadu_si128(i1 + 7);
+    v1 = _mm_loadu_si128(i2 + 7);
+    sad = _mm_sad_epu8(v0, v1);
+    acc = _mm_adds_epu16(acc, sad);
+
+    // This essential means sad = acc >> 64. We only care about the lower 16
+    // bits.
+    sad = _mm_shuffle_epi32(acc, 0xEE);
+    sad = _mm_adds_epu16(sad, acc);
+    int diff = _mm_cvtsi128_si32(sad);
+    if (diff)
+      return 1;
+    image1 += stride;
+    image2 += stride;
+  }
+  return 0;
+}
+
+}  // namespace remoting
-- 
cgit v1.1