1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
|
;
; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
;
; Use of this source code is governed by a BSD-style license and patent
; grant that can be found in the LICENSE file in the root of the source
; tree. 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"
%include "vp9_asm_enc_offsets.asm"
; void vp9_regular_quantize_b_sse4 | arg
; (BLOCK *b, | 0
; BLOCKD *d) | 1
global sym(vp9_regular_quantize_b_sse4)
sym(vp9_regular_quantize_b_sse4):
%if ABI_IS_32BIT
push rbp
mov rbp, rsp
GET_GOT rbx
push rdi
push rsi
ALIGN_STACK 16, rax
%define qcoeff 0 ; 32
%define stack_size 32
sub rsp, stack_size
%else
%ifidn __OUTPUT_FORMAT__,x64
SAVE_XMM 8, u
push rdi
push rsi
%endif
%endif
; end prolog
%if ABI_IS_32BIT
mov rdi, arg(0) ; BLOCK *b
mov rsi, arg(1) ; BLOCKD *d
%else
%ifidn __OUTPUT_FORMAT__,x64
mov rdi, rcx ; BLOCK *b
mov rsi, rdx ; BLOCKD *d
%else
;mov rdi, rdi ; BLOCK *b
;mov rsi, rsi ; BLOCKD *d
%endif
%endif
mov rax, [rdi + vp9_block_coeff]
mov rcx, [rdi + vp9_block_zbin]
mov rdx, [rdi + vp9_block_round]
movd xmm7, [rdi + vp9_block_zbin_extra]
; z
movdqa xmm0, [rax]
movdqa xmm1, [rax + 16]
; duplicate zbin_oq_value
pshuflw xmm7, xmm7, 0
punpcklwd xmm7, xmm7
movdqa xmm2, xmm0
movdqa xmm3, xmm1
; sz
psraw xmm0, 15
psraw xmm1, 15
; (z ^ sz)
pxor xmm2, xmm0
pxor xmm3, xmm1
; x = abs(z)
psubw xmm2, xmm0
psubw xmm3, xmm1
; zbin
movdqa xmm4, [rcx]
movdqa xmm5, [rcx + 16]
; *zbin_ptr + zbin_oq_value
paddw xmm4, xmm7
paddw xmm5, xmm7
movdqa xmm6, xmm2
movdqa xmm7, xmm3
; x - (*zbin_ptr + zbin_oq_value)
psubw xmm6, xmm4
psubw xmm7, xmm5
; round
movdqa xmm4, [rdx]
movdqa xmm5, [rdx + 16]
mov rax, [rdi + vp9_block_quant_shift]
mov rcx, [rdi + vp9_block_quant]
mov rdx, [rdi + vp9_block_zrun_zbin_boost]
; x + round
paddw xmm2, xmm4
paddw xmm3, xmm5
; quant
movdqa xmm4, [rcx]
movdqa xmm5, [rcx + 16]
; y = x * quant_ptr >> 16
pmulhw xmm4, xmm2
pmulhw xmm5, xmm3
; y += x
paddw xmm2, xmm4
paddw xmm3, xmm5
pxor xmm4, xmm4
%if ABI_IS_32BIT
movdqa [rsp + qcoeff], xmm4
movdqa [rsp + qcoeff + 16], xmm4
%else
pxor xmm8, xmm8
%endif
; quant_shift
movdqa xmm5, [rax]
; zrun_zbin_boost
mov rax, rdx
%macro ZIGZAG_LOOP 5
; x
pextrw ecx, %4, %2
; if (x >= zbin)
sub cx, WORD PTR[rdx] ; x - zbin
lea rdx, [rdx + 2] ; zbin_boost_ptr++
jl .rq_zigzag_loop_%1 ; x < zbin
pextrw edi, %3, %2 ; y
; downshift by quant_shift[rc]
pextrb ecx, xmm5, %1 ; quant_shift[rc]
sar edi, cl ; also sets Z bit
je .rq_zigzag_loop_%1 ; !y
%if ABI_IS_32BIT
mov WORD PTR[rsp + qcoeff + %1 *2], di
%else
pinsrw %5, edi, %2 ; qcoeff[rc]
%endif
mov rdx, rax ; reset to b->zrun_zbin_boost
.rq_zigzag_loop_%1:
%endmacro
; in vp9_default_zig_zag1d order: see vp9/common/vp9_entropy.c
ZIGZAG_LOOP 0, 0, xmm2, xmm6, xmm4
ZIGZAG_LOOP 1, 1, xmm2, xmm6, xmm4
ZIGZAG_LOOP 4, 4, xmm2, xmm6, xmm4
ZIGZAG_LOOP 8, 0, xmm3, xmm7, xmm8
ZIGZAG_LOOP 5, 5, xmm2, xmm6, xmm4
ZIGZAG_LOOP 2, 2, xmm2, xmm6, xmm4
ZIGZAG_LOOP 3, 3, xmm2, xmm6, xmm4
ZIGZAG_LOOP 6, 6, xmm2, xmm6, xmm4
ZIGZAG_LOOP 9, 1, xmm3, xmm7, xmm8
ZIGZAG_LOOP 12, 4, xmm3, xmm7, xmm8
ZIGZAG_LOOP 13, 5, xmm3, xmm7, xmm8
ZIGZAG_LOOP 10, 2, xmm3, xmm7, xmm8
ZIGZAG_LOOP 7, 7, xmm2, xmm6, xmm4
ZIGZAG_LOOP 11, 3, xmm3, xmm7, xmm8
ZIGZAG_LOOP 14, 6, xmm3, xmm7, xmm8
ZIGZAG_LOOP 15, 7, xmm3, xmm7, xmm8
mov rcx, [rsi + vp9_blockd_dequant]
mov rdi, [rsi + vp9_blockd_dqcoeff]
%if ABI_IS_32BIT
movdqa xmm4, [rsp + qcoeff]
movdqa xmm5, [rsp + qcoeff + 16]
%else
%define xmm5 xmm8
%endif
; y ^ sz
pxor xmm4, xmm0
pxor xmm5, xmm1
; x = (y ^ sz) - sz
psubw xmm4, xmm0
psubw xmm5, xmm1
; dequant
movdqa xmm0, [rcx]
movdqa xmm1, [rcx + 16]
mov rcx, [rsi + vp9_blockd_qcoeff]
pmullw xmm0, xmm4
pmullw xmm1, xmm5
; store qcoeff
movdqa [rcx], xmm4
movdqa [rcx + 16], xmm5
; store dqcoeff
movdqa [rdi], xmm0
movdqa [rdi + 16], xmm1
; select the last value (in zig_zag order) for EOB
pxor xmm6, xmm6
pcmpeqw xmm4, xmm6
pcmpeqw xmm5, xmm6
packsswb xmm4, xmm5
pshufb xmm4, [GLOBAL(zig_zag1d)]
pmovmskb edx, xmm4
xor rdi, rdi
mov eax, -1
xor dx, ax
bsr eax, edx
sub edi, edx
sar edi, 31
add eax, 1
and eax, edi
mov [rsi + vp9_blockd_eob], eax
; begin epilog
%if ABI_IS_32BIT
add rsp, stack_size
pop rsp
pop rsi
pop rdi
RESTORE_GOT
pop rbp
%else
%undef xmm5
%ifidn __OUTPUT_FORMAT__,x64
pop rsi
pop rdi
RESTORE_XMM
%endif
%endif
ret
SECTION_RODATA
align 16
; vp9/common/vp9_entropy.c: vp9_default_zig_zag1d
zig_zag1d:
db 0, 1, 4, 8, 5, 2, 3, 6, 9, 12, 13, 10, 7, 11, 14, 15
|
