From 7472021e53f9306b67ea1ba83fb5b232486537a3 Mon Sep 17 00:00:00 2001
From: "arphaxad.cy@gmail.com"
 <arphaxad.cy@gmail.com@16f28f9a-4ce2-e073-06de-1de4eb20be90>
Date: Mon, 31 Oct 2011 20:54:44 +0000
Subject: [PATCH] git-svn-id: http://libyuv.googlecode.com/svn/trunk@50
 16f28f9a-4ce2-e073-06de-1de4eb20be90

---
 source/rotate.cc | 130 ++++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 129 insertions(+), 1 deletion(-)

diff --git a/source/rotate.cc b/source/rotate.cc
index 14a5252f3..a1b05e898 100644
--- a/source/rotate.cc
+++ b/source/rotate.cc
@@ -367,7 +367,135 @@ static void TransposeWx8_SSSE3(const uint8* src, int src_stride,
 );
 }
 
-// TODO(fbarchard): Port to 32 bit
+#if defined (__i386__)
+extern "C" void TransposeUVWx8_SSE2(const uint8* src, int src_stride,
+                                    uint8* dst_a, int dst_stride_a,
+                                    uint8* dst_b, int dst_stride_b,
+                                    int w);
+  asm(
+    ".text\n"
+#if defined(OSX)
+    ".globl _TransposeUVWx8_SSE2\n"
+"_TransposeUVWx8_SSE2:\n"
+#else
+    ".global TransposeUVWx8_SSE2\n"
+"TransposeUVWx8_SSE2:\n"
+#endif
+    "push   %ebx\n"
+    "push   %esi\n"
+    "push   %edi\n"
+    "push   %ebp\n"
+    "mov    0x14(%esp),%eax\n"
+    "mov    0x18(%esp),%edi\n"
+    "mov    0x1c(%esp),%edx\n"
+    "mov    0x20(%esp),%esi\n"
+    "mov    0x24(%esp),%ebx\n"
+    "mov    0x28(%esp),%ebp\n"
+    "mov    %esp,%ecx\n"
+    "sub    $0x14,%esp\n"
+    "and    $0xfffffff0,%esp\n"
+    "mov    %ecx,0x10(%esp)\n"
+    "mov    0x2c(%ecx),%ecx\n"
+
+"1:"
+    "movdqa (%eax),%xmm0\n"
+    "movdqa (%eax,%edi,1),%xmm1\n"
+    "lea    (%eax,%edi,2),%eax\n"
+    "movdqa %xmm0,%xmm7\n"
+    "punpcklbw %xmm1,%xmm0\n"
+    "punpckhbw %xmm1,%xmm7\n"
+    "movdqa %xmm7,%xmm1\n"
+    "movdqa (%eax),%xmm2\n"
+    "movdqa (%eax,%edi,1),%xmm3\n"
+    "lea    (%eax,%edi,2),%eax\n"
+    "movdqa %xmm2,%xmm7\n"
+    "punpcklbw %xmm3,%xmm2\n"
+    "punpckhbw %xmm3,%xmm7\n"
+    "movdqa %xmm7,%xmm3\n"
+    "movdqa (%eax),%xmm4\n"
+    "movdqa (%eax,%edi,1),%xmm5\n"
+    "lea    (%eax,%edi,2),%eax\n"
+    "movdqa %xmm4,%xmm7\n"
+    "punpcklbw %xmm5,%xmm4\n"
+    "punpckhbw %xmm5,%xmm7\n"
+    "movdqa %xmm7,%xmm5\n"
+    "movdqa (%eax),%xmm6\n"
+    "movdqa (%eax,%edi,1),%xmm7\n"
+    "lea    (%eax,%edi,2),%eax\n"
+    "movdqa %xmm5,(%esp)\n"
+    "neg    %edi\n"
+    "movdqa %xmm6,%xmm5\n"
+    "punpcklbw %xmm7,%xmm6\n"
+    "punpckhbw %xmm7,%xmm5\n"
+    "movdqa %xmm5,%xmm7\n"
+    "lea    0x10(%eax,%edi,8),%eax\n"
+    "neg    %edi\n"
+    "movdqa %xmm0,%xmm5\n"
+    "punpcklwd %xmm2,%xmm0\n"
+    "punpckhwd %xmm2,%xmm5\n"
+    "movdqa %xmm5,%xmm2\n"
+    "movdqa %xmm1,%xmm5\n"
+    "punpcklwd %xmm3,%xmm1\n"
+    "punpckhwd %xmm3,%xmm5\n"
+    "movdqa %xmm5,%xmm3\n"
+    "movdqa %xmm4,%xmm5\n"
+    "punpcklwd %xmm6,%xmm4\n"
+    "punpckhwd %xmm6,%xmm5\n"
+    "movdqa %xmm5,%xmm6\n"
+    "movdqa (%esp),%xmm5\n"
+    "movdqa %xmm6,(%esp)\n"
+    "movdqa %xmm5,%xmm6\n"
+    "punpcklwd %xmm7,%xmm5\n"
+    "punpckhwd %xmm7,%xmm6\n"
+    "movdqa %xmm6,%xmm7\n"
+    "movdqa %xmm0,%xmm6\n"
+    "punpckldq %xmm4,%xmm0\n"
+    "punpckhdq %xmm4,%xmm6\n"
+    "movdqa %xmm6,%xmm4\n"
+    "movdqa (%esp),%xmm6\n"
+    "movlpd %xmm0,(%edx)\n"
+    "movhpd %xmm0,(%ebx)\n"
+    "movlpd %xmm4,(%edx,%esi,1)\n"
+    "lea    (%edx,%esi,2),%edx\n"
+    "movhpd %xmm4,(%ebx,%ebp,1)\n"
+    "lea    (%ebx,%ebp,2),%ebx\n"
+    "movdqa %xmm2,%xmm0\n"
+    "punpckldq %xmm6,%xmm2\n"
+    "movlpd %xmm2,(%edx)\n"
+    "movhpd %xmm2,(%ebx)\n"
+    "punpckhdq %xmm6,%xmm0\n"
+    "movlpd %xmm0,(%edx,%esi,1)\n"
+    "lea    (%edx,%esi,2),%edx\n"
+    "movhpd %xmm0,(%ebx,%ebp,1)\n"
+    "lea    (%ebx,%ebp,2),%ebx\n"
+    "movdqa %xmm1,%xmm0\n"
+    "punpckldq %xmm5,%xmm1\n"
+    "movlpd %xmm1,(%edx)\n"
+    "movhpd %xmm1,(%ebx)\n"
+    "punpckhdq %xmm5,%xmm0\n"
+    "movlpd %xmm0,(%edx,%esi,1)\n"
+    "lea    (%edx,%esi,2),%edx\n"
+    "movhpd %xmm0,(%ebx,%ebp,1)\n"
+    "lea    (%ebx,%ebp,2),%ebx\n"
+    "movdqa %xmm3,%xmm0\n"
+    "punpckldq %xmm7,%xmm3\n"
+    "movlpd %xmm3,(%edx)\n"
+    "movhpd %xmm3,(%ebx)\n"
+    "punpckhdq %xmm7,%xmm0\n"
+    "movlpd %xmm0,(%edx,%esi,1)\n"
+    "lea    (%edx,%esi,2),%edx\n"
+    "movhpd %xmm0,(%ebx,%ebp,1)\n"
+    "lea    (%ebx,%ebp,2),%ebx\n"
+    "sub    $0x8,%ecx\n"
+    "ja     1b\n"
+    "mov    0x10(%esp),%esp\n"
+    "pop    %ebp\n"
+    "pop    %edi\n"
+    "pop    %esi\n"
+    "pop    %ebx\n"
+    "ret\n"
+);
+
 #if defined (__x86_64__)
 #define HAS_TRANSPOSE_UVWX8_SSE2
 static void TransposeUVWx8_SSE2(const uint8* src, int src_stride,