Update libyuv to r1602 version to get best performance.android-cts_7.1_r1android-cts-7.1_r9android-cts-7.1_r8android-cts-7.1_r7android-cts-7.1_r6android-cts-7.1_r5android-cts-7.1_r4android-cts-7.1_r3android-cts-7.1_r29android-cts-7.1_r28android-cts-7.1_r27android-cts-7.1_r26android-cts-7.1_r25android-cts-7.1_r24android-cts-7.1_r23android-cts-7.1_r22android-cts-7.1_r21android-cts-7.1_r20android-cts-7.1_r2android-cts-7.1_r19android-cts-7.1_r18android-cts-7.1_r17android-cts-7.1_r16android-cts-7.1_r15android-cts-7.1_r14android-cts-7.1_r13android-cts-7.1_r12android-cts-7.1_r11android-cts-7.1_r10android-cts-7.1_r1android-7.1.2_r9android-7.1.2_r8android-7.1.2_r6android-7.1.2_r5android-7.1.2_r4android-7.1.2_r39android-7.1.2_r38android-7.1.2_r37android-7.1.2_r36android-7.1.2_r33android-7.1.2_r32android-7.1.2_r30android-7.1.2_r3android-7.1.2_r29android-7.1.2_r28android-7.1.2_r27android-7.1.2_r25android-7.1.2_r24android-7.1.2_r23android-7.1.2_r2android-7.1.2_r19android-7.1.2_r18android-7.1.2_r17android-7.1.2_r16android-7.1.2_r15android-7.1.2_r14android-7.1.2_r13android-7.1.2_r12android-7.1.2_r11android-7.1.2_r10android-7.1.2_r1android-7.1.1_r9android-7.1.1_r8android-7.1.1_r7android-7.1.1_r61android-7.1.1_r60android-7.1.1_r6android-7.1.1_r59android-7.1.1_r58android-7.1.1_r57android-7.1.1_r56android-7.1.1_r55android-7.1.1_r54android-7.1.1_r53android-7.1.1_r52android-7.1.1_r51android-7.1.1_r50android-7.1.1_r49android-7.1.1_r48android-7.1.1_r47android-7.1.1_r46android-7.1.1_r45android-7.1.1_r44android-7.1.1_r43android-7.1.1_r42android-7.1.1_r41android-7.1.1_r40android-7.1.1_r4android-7.1.1_r39android-7.1.1_r38android-7.1.1_r35android-7.1.1_r33android-7.1.1_r32android-7.1.1_r31android-7.1.1_r3android-7.1.1_r28android-7.1.1_r27android-7.1.1_r26android-7.1.1_r25android-7.1.1_r24android-7.1.1_r23android-7.1.1_r22android-7.1.1_r21android-7.1.1_r20android-7.1.1_r2android-7.1.1_r17android-7.1.1_r16android-7.1.1_r15android-7.1.1_r14android-7.1.1_r13android-7.1.1_r12android-7.1.1_r11android-7.1.1_r10android-7.1.1_r1android-7.1.0_r7android-7.1.0_r6android-7.1.0_r5android-7.1.0_r4android-7.1.0_r3android-7.1.0_r2android-7.1.0_r1nougat-mr2.3-releasenougat-mr2.2-releasenougat-mr2.1-releasenougat-mr2-security-releasenougat-mr2-releasenougat-mr2-pixel-releasenougat-mr2-devnougat-mr1.8-releasenougat-mr1.7-releasenougat-mr1.6-releasenougat-mr1.5-releasenougat-mr1.4-releasenougat-mr1.3-releasenougat-mr1.2-releasenougat-mr1.1-releasenougat-mr1-volantis-releasenougat-mr1-security-releasenougat-mr1-releasenougat-mr1-flounder-releasenougat-mr1-devnougat-mr1-cts-releasenougat-dr1-release

Bug: 29870647

Change-Id: I8ec9fab7f55765fa33ebe7ba1c7ad2147f418de2

1 files changed, 151 insertions, 0 deletions

diff --git a/files/source/compare_gcc.cc b/files/source/compare_gcc.cc
new file mode 100644
index 00000000..1b83edb1
--- /dev/null
+++ b/files/source/compare_gcc.cc
@@ -0,0 +1,151 @@
+/*
+ *  Copyright 2012 The LibYuv 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 "libyuv/basic_types.h"
+
+#include "libyuv/compare_row.h"
+#include "libyuv/row.h"
+
+#ifdef __cplusplus
+namespace libyuv {
+extern "C" {
+#endif
+
+// This module is for GCC x86 and x64.
+#if !defined(LIBYUV_DISABLE_X86) && \
+    (defined(__x86_64__) || (defined(__i386__) && !defined(_MSC_VER)))
+
+uint32 SumSquareError_SSE2(const uint8* src_a, const uint8* src_b, int count) {
+  uint32 sse;
+  asm volatile (
+    "pxor      %%xmm0,%%xmm0                   \n"
+    "pxor      %%xmm5,%%xmm5                   \n"
+    LABELALIGN
+  "1:                                          \n"
+    "movdqu    " MEMACCESS(0) ",%%xmm1         \n"
+    "lea       " MEMLEA(0x10, 0) ",%0          \n"
+    "movdqu    " MEMACCESS(1) ",%%xmm2         \n"
+    "lea       " MEMLEA(0x10, 1) ",%1          \n"
+    "movdqa    %%xmm1,%%xmm3                   \n"
+    "psubusb   %%xmm2,%%xmm1                   \n"
+    "psubusb   %%xmm3,%%xmm2                   \n"
+    "por       %%xmm2,%%xmm1                   \n"
+    "movdqa    %%xmm1,%%xmm2                   \n"
+    "punpcklbw %%xmm5,%%xmm1                   \n"
+    "punpckhbw %%xmm5,%%xmm2                   \n"
+    "pmaddwd   %%xmm1,%%xmm1                   \n"
+    "pmaddwd   %%xmm2,%%xmm2                   \n"
+    "paddd     %%xmm1,%%xmm0                   \n"
+    "paddd     %%xmm2,%%xmm0                   \n"
+    "sub       $0x10,%2                        \n"
+    "jg        1b                              \n"
+
+    "pshufd    $0xee,%%xmm0,%%xmm1             \n"
+    "paddd     %%xmm1,%%xmm0                   \n"
+    "pshufd    $0x1,%%xmm0,%%xmm1              \n"
+    "paddd     %%xmm1,%%xmm0                   \n"
+    "movd      %%xmm0,%3                       \n"
+
+  : "+r"(src_a),      // %0
+    "+r"(src_b),      // %1
+    "+r"(count),      // %2
+    "=g"(sse)         // %3
+  :: "memory", "cc", "xmm0", "xmm1", "xmm2", "xmm3", "xmm5"
+  );
+  return sse;
+}
+
+static uvec32 kHash16x33 = { 0x92d9e201, 0, 0, 0 };  // 33 ^ 16
+static uvec32 kHashMul0 = {
+  0x0c3525e1,  // 33 ^ 15
+  0xa3476dc1,  // 33 ^ 14
+  0x3b4039a1,  // 33 ^ 13
+  0x4f5f0981,  // 33 ^ 12
+};
+static uvec32 kHashMul1 = {
+  0x30f35d61,  // 33 ^ 11
+  0x855cb541,  // 33 ^ 10
+  0x040a9121,  // 33 ^ 9
+  0x747c7101,  // 33 ^ 8
+};
+static uvec32 kHashMul2 = {
+  0xec41d4e1,  // 33 ^ 7
+  0x4cfa3cc1,  // 33 ^ 6
+  0x025528a1,  // 33 ^ 5
+  0x00121881,  // 33 ^ 4
+};
+static uvec32 kHashMul3 = {
+  0x00008c61,  // 33 ^ 3
+  0x00000441,  // 33 ^ 2
+  0x00000021,  // 33 ^ 1
+  0x00000001,  // 33 ^ 0
+};
+
+uint32 HashDjb2_SSE41(const uint8* src, int count, uint32 seed) {
+  uint32 hash;
+  asm volatile (
+    "movd      %2,%%xmm0                       \n"
+    "pxor      %%xmm7,%%xmm7                   \n"
+    "movdqa    %4,%%xmm6                       \n"
+    LABELALIGN
+  "1:                                          \n"
+    "movdqu    " MEMACCESS(0) ",%%xmm1         \n"
+    "lea       " MEMLEA(0x10, 0) ",%0          \n"
+    "pmulld    %%xmm6,%%xmm0                   \n"
+    "movdqa    %5,%%xmm5                       \n"
+    "movdqa    %%xmm1,%%xmm2                   \n"
+    "punpcklbw %%xmm7,%%xmm2                   \n"
+    "movdqa    %%xmm2,%%xmm3                   \n"
+    "punpcklwd %%xmm7,%%xmm3                   \n"
+    "pmulld    %%xmm5,%%xmm3                   \n"
+    "movdqa    %6,%%xmm5                       \n"
+    "movdqa    %%xmm2,%%xmm4                   \n"
+    "punpckhwd %%xmm7,%%xmm4                   \n"
+    "pmulld    %%xmm5,%%xmm4                   \n"
+    "movdqa    %7,%%xmm5                       \n"
+    "punpckhbw %%xmm7,%%xmm1                   \n"
+    "movdqa    %%xmm1,%%xmm2                   \n"
+    "punpcklwd %%xmm7,%%xmm2                   \n"
+    "pmulld    %%xmm5,%%xmm2                   \n"
+    "movdqa    %8,%%xmm5                       \n"
+    "punpckhwd %%xmm7,%%xmm1                   \n"
+    "pmulld    %%xmm5,%%xmm1                   \n"
+    "paddd     %%xmm4,%%xmm3                   \n"
+    "paddd     %%xmm2,%%xmm1                   \n"
+    "paddd     %%xmm3,%%xmm1                   \n"
+    "pshufd    $0xe,%%xmm1,%%xmm2              \n"
+    "paddd     %%xmm2,%%xmm1                   \n"
+    "pshufd    $0x1,%%xmm1,%%xmm2              \n"
+    "paddd     %%xmm2,%%xmm1                   \n"
+    "paddd     %%xmm1,%%xmm0                   \n"
+    "sub       $0x10,%1                        \n"
+    "jg        1b                              \n"
+    "movd      %%xmm0,%3                       \n"
+  : "+r"(src),        // %0
+    "+r"(count),      // %1
+    "+rm"(seed),      // %2
+    "=g"(hash)        // %3
+  : "m"(kHash16x33),  // %4
+    "m"(kHashMul0),   // %5
+    "m"(kHashMul1),   // %6
+    "m"(kHashMul2),   // %7
+    "m"(kHashMul3)    // %8
+  : "memory", "cc"
+    , "xmm0", "xmm1", "xmm2", "xmm3", "xmm4", "xmm5", "xmm6", "xmm7"
+  );
+  return hash;
+}
+#endif  // defined(__x86_64__) || (defined(__i386__) && !defined(__pic__)))
+
+#ifdef __cplusplus
+}  // extern "C"
+}  // namespace libyuv
+#endif
+