summaryrefslogtreecommitdiff
path: root/vp9/common/x86/vp9_iwalsh_mmx.asm
diff options
context:
space:
mode:
authorJohn Koleszar <jkoleszar@google.com>2012-11-27 13:59:17 -0800
committerJohn Koleszar <jkoleszar@google.com>2012-11-27 14:12:30 -0800
commitfcccbcbb395ce4cf31b54ce1245cc28e5e3ef4c1 (patch)
tree68e128e48e3f5ab1de1c163fa3a12ea47f5d8d51 /vp9/common/x86/vp9_iwalsh_mmx.asm
parent3bf7b131c8ebb6b4d63a8b70d69066dcbc4ed896 (diff)
downloadlibvpx-fcccbcbb395ce4cf31b54ce1245cc28e5e3ef4c1.tar
libvpx-fcccbcbb395ce4cf31b54ce1245cc28e5e3ef4c1.tar.gz
libvpx-fcccbcbb395ce4cf31b54ce1245cc28e5e3ef4c1.tar.bz2
libvpx-fcccbcbb395ce4cf31b54ce1245cc28e5e3ef4c1.zip
Add vp9_ prefix to all vp9 files
Support for gyp which doesn't support multiple objects in the same static library having the same basename. Change-Id: Ib947eefbaf68f8b177a796d23f875ccdfa6bc9dc
Diffstat (limited to 'vp9/common/x86/vp9_iwalsh_mmx.asm')
-rw-r--r--vp9/common/x86/vp9_iwalsh_mmx.asm173
1 files changed, 173 insertions, 0 deletions
diff --git a/vp9/common/x86/vp9_iwalsh_mmx.asm b/vp9/common/x86/vp9_iwalsh_mmx.asm
new file mode 100644
index 000000000..6b276b95a
--- /dev/null
+++ b/vp9/common/x86/vp9_iwalsh_mmx.asm
@@ -0,0 +1,173 @@
+;
+; Copyright (c) 2010 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.
+;
+
+
+%include "vpx_ports/x86_abi_support.asm"
+
+;void vp9_short_inv_walsh4x4_1_mmx(short *input, short *output)
+global sym(vp9_short_inv_walsh4x4_1_mmx)
+sym(vp9_short_inv_walsh4x4_1_mmx):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 2
+ push rsi
+ push rdi
+ ; end prolog
+
+ mov rsi, arg(0)
+ mov rax, 3
+
+ mov rdi, arg(1)
+ add rax, [rsi] ;input[0] + 3
+
+ movd mm0, eax
+
+ punpcklwd mm0, mm0 ;x x val val
+
+ punpckldq mm0, mm0 ;val val val val
+
+ psraw mm0, 3 ;(input[0] + 3) >> 3
+
+ movq [rdi + 0], mm0
+ movq [rdi + 8], mm0
+ movq [rdi + 16], mm0
+ movq [rdi + 24], mm0
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+
+;void vp9_short_inv_walsh4x4_mmx(short *input, short *output)
+global sym(vp9_short_inv_walsh4x4_mmx)
+sym(vp9_short_inv_walsh4x4_mmx):
+ push rbp
+ mov rbp, rsp
+ SHADOW_ARGS_TO_STACK 2
+ push rsi
+ push rdi
+ ; end prolog
+
+ mov rax, 3
+ mov rsi, arg(0)
+ mov rdi, arg(1)
+ shl rax, 16
+
+ movq mm0, [rsi + 0] ;ip[0]
+ movq mm1, [rsi + 8] ;ip[4]
+ or rax, 3 ;00030003h
+
+ movq mm2, [rsi + 16] ;ip[8]
+ movq mm3, [rsi + 24] ;ip[12]
+
+ movq mm7, rax
+ movq mm4, mm0
+
+ punpcklwd mm7, mm7 ;0003000300030003h
+ movq mm5, mm1
+
+ paddw mm4, mm3 ;ip[0] + ip[12] aka al
+ paddw mm5, mm2 ;ip[4] + ip[8] aka bl
+
+ movq mm6, mm4 ;temp al
+
+ paddw mm4, mm5 ;al + bl
+ psubw mm6, mm5 ;al - bl
+
+ psubw mm0, mm3 ;ip[0] - ip[12] aka d1
+ psubw mm1, mm2 ;ip[4] - ip[8] aka c1
+
+ movq mm5, mm0 ;temp dl
+
+ paddw mm0, mm1 ;dl + cl
+ psubw mm5, mm1 ;dl - cl
+
+ ; 03 02 01 00
+ ; 13 12 11 10
+ ; 23 22 21 20
+ ; 33 32 31 30
+
+ movq mm3, mm4 ; 03 02 01 00
+ punpcklwd mm4, mm0 ; 11 01 10 00
+ punpckhwd mm3, mm0 ; 13 03 12 02
+
+ movq mm1, mm6 ; 23 22 21 20
+ punpcklwd mm6, mm5 ; 31 21 30 20
+ punpckhwd mm1, mm5 ; 33 23 32 22
+
+ movq mm0, mm4 ; 11 01 10 00
+ movq mm2, mm3 ; 13 03 12 02
+
+ punpckldq mm0, mm6 ; 30 20 10 00 aka ip[0]
+ punpckhdq mm4, mm6 ; 31 21 11 01 aka ip[4]
+
+ punpckldq mm2, mm1 ; 32 22 12 02 aka ip[8]
+ punpckhdq mm3, mm1 ; 33 23 13 03 aka ip[12]
+;~~~~~~~~~~~~~~~~~~~~~
+ movq mm1, mm0
+ movq mm5, mm4
+
+ paddw mm1, mm3 ;ip[0] + ip[12] aka al
+ paddw mm5, mm2 ;ip[4] + ip[8] aka bl
+
+ movq mm6, mm1 ;temp al
+
+ paddw mm1, mm5 ;al + bl
+ psubw mm6, mm5 ;al - bl
+
+ psubw mm0, mm3 ;ip[0] - ip[12] aka d1
+ psubw mm4, mm2 ;ip[4] - ip[8] aka c1
+
+ movq mm5, mm0 ;temp dl
+
+ paddw mm0, mm4 ;dl + cl
+ psubw mm5, mm4 ;dl - cl
+;~~~~~~~~~~~~~~~~~~~~~
+ movq mm3, mm1 ; 03 02 01 00
+ punpcklwd mm1, mm0 ; 11 01 10 00
+ punpckhwd mm3, mm0 ; 13 03 12 02
+
+ movq mm4, mm6 ; 23 22 21 20
+ punpcklwd mm6, mm5 ; 31 21 30 20
+ punpckhwd mm4, mm5 ; 33 23 32 22
+
+ movq mm0, mm1 ; 11 01 10 00
+ movq mm2, mm3 ; 13 03 12 02
+
+ punpckldq mm0, mm6 ; 30 20 10 00 aka ip[0]
+ punpckhdq mm1, mm6 ; 31 21 11 01 aka ip[4]
+
+ punpckldq mm2, mm4 ; 32 22 12 02 aka ip[8]
+ punpckhdq mm3, mm4 ; 33 23 13 03 aka ip[12]
+
+ paddw mm0, mm7
+ paddw mm1, mm7
+ paddw mm2, mm7
+ paddw mm3, mm7
+
+ psraw mm0, 3
+ psraw mm1, 3
+ psraw mm2, 3
+ psraw mm3, 3
+
+ movq [rdi + 0], mm0
+ movq [rdi + 8], mm1
+ movq [rdi + 16], mm2
+ movq [rdi + 24], mm3
+
+ ; begin epilog
+ pop rdi
+ pop rsi
+ UNSHADOW_ARGS
+ pop rbp
+ ret
+