From 35508d09796c8186c5b94992fdbaed79a6314f37 Mon Sep 17 00:00:00 2001
From: "fbarchard@google.com" <fbarchard@google.com>
Date: Thu, 13 Nov 2014 23:11:10 +0000
Subject: [PATCH] Mirror_AVX2 ported to GCC. BUG=269 TESTED=try bots
 R=harryjin@google.com

Review URL: https://webrtc-codereview.appspot.com/32079004

git-svn-id: http://libyuv.googlecode.com/svn/trunk@1164 16f28f9a-4ce2-e073-06de-1de4eb20be90
---
 include/libyuv/row.h |  4 ++--
 source/row_posix.cc  | 31 +++++++++++++++++++++++++++++++
 source/row_win.cc    |  8 ++------
 3 files changed, 35 insertions(+), 8 deletions(-)

diff --git a/include/libyuv/row.h b/include/libyuv/row.h
index 941be842c..ffd837030 100644
--- a/include/libyuv/row.h
+++ b/include/libyuv/row.h
@@ -199,6 +199,7 @@ extern "C" {
 #define HAS_UYVYTOUVROW_AVX2
 #define HAS_SPLITUVROW_AVX2
 #define HAS_MERGEUVROW_AVX2
+#define HAS_MIRRORROW_AVX2
 
 // Effects:
 #define HAS_ARGBADDROW_AVX2
@@ -206,6 +207,7 @@ extern "C" {
 #define HAS_ARGBMULTIPLYROW_AVX2
 #define HAS_ARGBATTENUATEROW_AVX2
 #define HAS_ARGBUNATTENUATEROW_AVX2
+#define HAS_ARGBMIRRORROW_AVX2
 #endif
 
 // The following are require VS2012.
@@ -218,8 +220,6 @@ extern "C" {
 #define HAS_I422TORGBAROW_AVX2
 #define HAS_I422TOABGRROW_AVX2
 #define HAS_INTERPOLATEROW_AVX2
-#define HAS_MIRRORROW_AVX2
-#define HAS_ARGBMIRRORROW_AVX2
 #endif  // defined(VISUALC_HAS_AVX2)
 
 // The following are Yasm x86 only:
diff --git a/source/row_posix.cc b/source/row_posix.cc
index 77047a28c..18927600b 100644
--- a/source/row_posix.cc
+++ b/source/row_posix.cc
@@ -2207,6 +2207,37 @@ void MirrorRow_SSSE3(const uint8* src, uint8* dst, int width) {
 }
 #endif  // HAS_MIRRORROW_SSSE3
 
+#ifdef HAS_MIRRORROW_AVX2
+void MirrorRow_AVX2(const uint8* src, uint8* dst, int width) {
+  intptr_t temp_width = (intptr_t)(width);
+  asm volatile (
+    "vbroadcastf128 %3,%%ymm5                  \n"
+    "lea       " MEMLEA(-0x20,0) ",%0          \n"
+    LABELALIGN
+  "1:                                          \n"
+    MEMOPREG(vmovdqu,0x00,0,2,1,ymm0)          //  vmovdqu (%0,%2),%%ymm0
+    "vpshufb    %%ymm5,%%ymm0,%%ymm0           \n"
+    "vpermq     $0x4e,%%ymm0,%%ymm0            \n"
+    "sub       $0x20,%2                        \n"
+    "vmovdqu    %%ymm0," MEMACCESS(1) "        \n"
+    "lea       " MEMLEA(0x20,1) ",%1           \n"
+    "jg        1b                              \n"
+    "vzeroupper                                \n"
+  : "+r"(src),  // %0
+    "+r"(dst),  // %1
+    "+r"(temp_width)  // %2
+  : "m"(kShuffleMirror) // %3
+  : "memory", "cc"
+#if defined(__native_client__) && defined(__x86_64__)
+    , "r14"
+#endif
+#if defined(__SSE2__)
+    , "xmm0", "xmm5"
+#endif
+  );
+}
+#endif  // HAS_MIRRORROW_AVX2
+
 #ifdef HAS_MIRRORROW_SSE2
 void MirrorRow_SSE2(const uint8* src, uint8* dst, int width) {
   intptr_t temp_width = (intptr_t)(width);
diff --git a/source/row_win.cc b/source/row_win.cc
index d07dc620e..6edc6be77 100644
--- a/source/row_win.cc
+++ b/source/row_win.cc
@@ -2398,6 +2398,7 @@ static const uvec8 kShuffleMirror = {
   15u, 14u, 13u, 12u, 11u, 10u, 9u, 8u, 7u, 6u, 5u, 4u, 3u, 2u, 1u, 0u
 };
 
+// TODO(fbarchard): Replace lea with -16 offset.
 __declspec(naked) __declspec(align(16))
 void MirrorRow_SSSE3(const uint8* src, uint8* dst, int width) {
   __asm {
@@ -2421,18 +2422,13 @@ void MirrorRow_SSSE3(const uint8* src, uint8* dst, int width) {
 #endif  // HAS_MIRRORROW_SSSE3
 
 #ifdef HAS_MIRRORROW_AVX2
-// Shuffle table for reversing the bytes.
-static const uvec8 kShuffleMirror_AVX2 = {
-  15u, 14u, 13u, 12u, 11u, 10u, 9u, 8u, 7u, 6u, 5u, 4u, 3u, 2u, 1u, 0u
-};
-
 __declspec(naked) __declspec(align(16))
 void MirrorRow_AVX2(const uint8* src, uint8* dst, int width) {
   __asm {
     mov       eax, [esp + 4]   // src
     mov       edx, [esp + 8]   // dst
     mov       ecx, [esp + 12]  // width
-    vbroadcastf128 ymm5, kShuffleMirror_AVX2
+    vbroadcastf128 ymm5, kShuffleMirror
     lea       eax, [eax - 32]
 
     align      4