modify SAVE_XMM for potential 64bit use
[libvpx.git] / vp8 / encoder / x86 / ssim_opt.asm
blobd5d267a69847ba989b6cb4aaa17783043835df5d
2 ; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
4 ; Use of this source code is governed by a BSD-style license
5 ; that can be found in the LICENSE file in the root of the source
6 ; tree. An additional intellectual property rights grant can be found
7 ; in the file PATENTS. All contributing project authors may
8 ; be found in the AUTHORS file in the root of the source tree.
11 %include "vpx_ports/x86_abi_support.asm"
13 ; tabulate_ssim - sums sum_s,sum_r,sum_sq_s,sum_sq_r, sum_sxr
14 %macro TABULATE_SSIM 0
15 paddusw xmm15, xmm3 ; sum_s
16 paddusw xmm14, xmm4 ; sum_r
17 movdqa xmm1, xmm3
18 pmaddwd xmm1, xmm1
19 paddd xmm13, xmm1 ; sum_sq_s
20 movdqa xmm2, xmm4
21 pmaddwd xmm2, xmm2
22 paddd xmm12, xmm2 ; sum_sq_r
23 pmaddwd xmm3, xmm4
24 paddd xmm11, xmm3 ; sum_sxr
25 %endmacro
27 ; Sum across the register %1 starting with q words
28 %macro SUM_ACROSS_Q 1
29 movdqa xmm2,%1
30 punpckldq %1,xmm0
31 punpckhdq xmm2,xmm0
32 paddq %1,xmm2
33 movdqa xmm2,%1
34 punpcklqdq %1,xmm0
35 punpckhqdq xmm2,xmm0
36 paddq %1,xmm2
37 %endmacro
39 ; Sum across the register %1 starting with q words
40 %macro SUM_ACROSS_W 1
41 movdqa xmm1, %1
42 punpcklwd %1,xmm0
43 punpckhwd xmm1,xmm0
44 paddd %1, xmm1
45 SUM_ACROSS_Q %1
46 %endmacro
47 ;void ssim_parms_sse3(
48 ; unsigned char *s,
49 ; int sp,
50 ; unsigned char *r,
51 ; int rp
52 ; unsigned long *sum_s,
53 ; unsigned long *sum_r,
54 ; unsigned long *sum_sq_s,
55 ; unsigned long *sum_sq_r,
56 ; unsigned long *sum_sxr);
58 ; TODO: Use parm passing through structure, probably don't need the pxors
59 ; ( calling app will initialize to 0 ) could easily fit everything in sse2
60 ; without too much hastle, and can probably do better estimates with psadw
61 ; or pavgb At this point this is just meant to be first pass for calculating
62 ; all the parms needed for 16x16 ssim so we can play with dssim as distortion
63 ; in mode selection code.
64 global sym(vp8_ssim_parms_16x16_sse3)
65 sym(vp8_ssim_parms_16x16_sse3):
66 push rbp
67 mov rbp, rsp
68 SHADOW_ARGS_TO_STACK 9
69 SAVE_XMM 15
70 push rsi
71 push rdi
72 ; end prolog
74 mov rsi, arg(0) ;s
75 mov rcx, arg(1) ;sp
76 mov rdi, arg(2) ;r
77 mov rax, arg(3) ;rp
79 pxor xmm0, xmm0
80 pxor xmm15,xmm15 ;sum_s
81 pxor xmm14,xmm14 ;sum_r
82 pxor xmm13,xmm13 ;sum_sq_s
83 pxor xmm12,xmm12 ;sum_sq_r
84 pxor xmm11,xmm11 ;sum_sxr
86 mov rdx, 16 ;row counter
87 NextRow:
89 ;grab source and reference pixels
90 movdqu xmm5, [rsi]
91 movdqu xmm6, [rdi]
92 movdqa xmm3, xmm5
93 movdqa xmm4, xmm6
94 punpckhbw xmm3, xmm0 ; high_s
95 punpckhbw xmm4, xmm0 ; high_r
97 TABULATE_SSIM
99 movdqa xmm3, xmm5
100 movdqa xmm4, xmm6
101 punpcklbw xmm3, xmm0 ; low_s
102 punpcklbw xmm4, xmm0 ; low_r
104 TABULATE_SSIM
106 add rsi, rcx ; next s row
107 add rdi, rax ; next r row
109 dec rdx ; counter
110 jnz NextRow
112 SUM_ACROSS_W xmm15
113 SUM_ACROSS_W xmm14
114 SUM_ACROSS_Q xmm13
115 SUM_ACROSS_Q xmm12
116 SUM_ACROSS_Q xmm11
118 mov rdi,arg(4)
119 movd [rdi], xmm15;
120 mov rdi,arg(5)
121 movd [rdi], xmm14;
122 mov rdi,arg(6)
123 movd [rdi], xmm13;
124 mov rdi,arg(7)
125 movd [rdi], xmm12;
126 mov rdi,arg(8)
127 movd [rdi], xmm11;
129 ; begin epilog
130 pop rdi
131 pop rsi
132 RESTORE_XMM
133 UNSHADOW_ARGS
134 pop rbp
137 ;void ssim_parms_sse3(
138 ; unsigned char *s,
139 ; int sp,
140 ; unsigned char *r,
141 ; int rp
142 ; unsigned long *sum_s,
143 ; unsigned long *sum_r,
144 ; unsigned long *sum_sq_s,
145 ; unsigned long *sum_sq_r,
146 ; unsigned long *sum_sxr);
148 ; TODO: Use parm passing through structure, probably don't need the pxors
149 ; ( calling app will initialize to 0 ) could easily fit everything in sse2
150 ; without too much hastle, and can probably do better estimates with psadw
151 ; or pavgb At this point this is just meant to be first pass for calculating
152 ; all the parms needed for 16x16 ssim so we can play with dssim as distortion
153 ; in mode selection code.
154 global sym(vp8_ssim_parms_8x8_sse3)
155 sym(vp8_ssim_parms_8x8_sse3):
156 push rbp
157 mov rbp, rsp
158 SHADOW_ARGS_TO_STACK 9
159 SAVE_XMM 15
160 push rsi
161 push rdi
162 ; end prolog
164 mov rsi, arg(0) ;s
165 mov rcx, arg(1) ;sp
166 mov rdi, arg(2) ;r
167 mov rax, arg(3) ;rp
169 pxor xmm0, xmm0
170 pxor xmm15,xmm15 ;sum_s
171 pxor xmm14,xmm14 ;sum_r
172 pxor xmm13,xmm13 ;sum_sq_s
173 pxor xmm12,xmm12 ;sum_sq_r
174 pxor xmm11,xmm11 ;sum_sxr
176 mov rdx, 8 ;row counter
177 NextRow2:
179 ;grab source and reference pixels
180 movq xmm3, [rsi]
181 movq xmm4, [rdi]
182 punpcklbw xmm3, xmm0 ; low_s
183 punpcklbw xmm4, xmm0 ; low_r
185 TABULATE_SSIM
187 add rsi, rcx ; next s row
188 add rdi, rax ; next r row
190 dec rdx ; counter
191 jnz NextRow2
193 SUM_ACROSS_W xmm15
194 SUM_ACROSS_W xmm14
195 SUM_ACROSS_Q xmm13
196 SUM_ACROSS_Q xmm12
197 SUM_ACROSS_Q xmm11
199 mov rdi,arg(4)
200 movd [rdi], xmm15;
201 mov rdi,arg(5)
202 movd [rdi], xmm14;
203 mov rdi,arg(6)
204 movd [rdi], xmm13;
205 mov rdi,arg(7)
206 movd [rdi], xmm12;
207 mov rdi,arg(8)
208 movd [rdi], xmm11;
210 ; begin epilog
211 pop rdi
212 pop rsi
213 RESTORE_XMM
214 UNSHADOW_ARGS
215 pop rbp