mfqe_sse2.asm 7.14 KB
Newer Older
Johann's avatar
Johann committed
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281
;
;  Copyright (c) 2012 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 vp8_filter_by_weight16x16_sse2
;(
;    unsigned char *src,
;    int            src_stride,
;    unsigned char *dst,
;    int            dst_stride,
;    int            src_weight
;)
global sym(vp8_filter_by_weight16x16_sse2)
sym(vp8_filter_by_weight16x16_sse2):
    push        rbp
    mov         rbp, rsp
    SHADOW_ARGS_TO_STACK 5
    SAVE_XMM 6
    GET_GOT     rbx
    push        rsi
    push        rdi
    ; end prolog

    movd        xmm0, arg(4)                ; src_weight
    pshuflw     xmm0, xmm0, 0x0             ; replicate to all low words
    punpcklqdq  xmm0, xmm0                  ; replicate to all hi words

    movdqa      xmm1, [GLOBAL(tMFQE)]
    psubw       xmm1, xmm0                  ; dst_weight

    mov         rax, arg(0)                 ; src
    mov         rsi, arg(1)                 ; src_stride
    mov         rdx, arg(2)                 ; dst
    mov         rdi, arg(3)                 ; dst_stride

    mov         rcx, 16                     ; loop count
    pxor        xmm6, xmm6

.combine
    movdqa      xmm2, [rax]
    movdqa      xmm4, [rdx]
    add         rax, rsi

    ; src * src_weight
    movdqa      xmm3, xmm2
    punpcklbw   xmm2, xmm6
    punpckhbw   xmm3, xmm6
    pmullw      xmm2, xmm0
    pmullw      xmm3, xmm0

    ; dst * dst_weight
    movdqa      xmm5, xmm4
    punpcklbw   xmm4, xmm6
    punpckhbw   xmm5, xmm6
    pmullw      xmm4, xmm1
    pmullw      xmm5, xmm1

    ; sum, round and shift
    paddw       xmm2, xmm4
    paddw       xmm3, xmm5
    paddw       xmm2, [GLOBAL(tMFQE_round)]
    paddw       xmm3, [GLOBAL(tMFQE_round)]
    psrlw       xmm2, 4
    psrlw       xmm3, 4

    packuswb    xmm2, xmm3
    movdqa      [rdx], xmm2
    add         rdx, rdi

    dec         rcx
    jnz         .combine

    ; begin epilog
    pop         rdi
    pop         rsi
    RESTORE_GOT
    RESTORE_XMM
    UNSHADOW_ARGS
    pop         rbp

    ret

;void vp8_filter_by_weight8x8_sse2
;(
;    unsigned char *src,
;    int            src_stride,
;    unsigned char *dst,
;    int            dst_stride,
;    int            src_weight
;)
global sym(vp8_filter_by_weight8x8_sse2)
sym(vp8_filter_by_weight8x8_sse2):
    push        rbp
    mov         rbp, rsp
    SHADOW_ARGS_TO_STACK 5
    GET_GOT     rbx
    push        rsi
    push        rdi
    ; end prolog

    movd        xmm0, arg(4)                ; src_weight
    pshuflw     xmm0, xmm0, 0x0             ; replicate to all low words
    punpcklqdq  xmm0, xmm0                  ; replicate to all hi words

    movdqa      xmm1, [GLOBAL(tMFQE)]
    psubw       xmm1, xmm0                  ; dst_weight

    mov         rax, arg(0)                 ; src
    mov         rsi, arg(1)                 ; src_stride
    mov         rdx, arg(2)                 ; dst
    mov         rdi, arg(3)                 ; dst_stride

    mov         rcx, 8                      ; loop count
    pxor        xmm4, xmm4

.combine
    movq        xmm2, [rax]
    movq        xmm3, [rdx]
    add         rax, rsi

    ; src * src_weight
    punpcklbw   xmm2, xmm4
    pmullw      xmm2, xmm0

    ; dst * dst_weight
    punpcklbw   xmm3, xmm4
    pmullw      xmm3, xmm1

    ; sum, round and shift
    paddw       xmm2, xmm3
    paddw       xmm2, [GLOBAL(tMFQE_round)]
    psrlw       xmm2, 4

    packuswb    xmm2, xmm4
    movq        [rdx], xmm2
    add         rdx, rdi

    dec         rcx
    jnz         .combine

    ; begin epilog
    pop         rdi
    pop         rsi
    RESTORE_GOT
    UNSHADOW_ARGS
    pop         rbp

    ret

;void vp8_variance_and_sad_16x16_sse2 | arg
;(
;    unsigned char *src1,          0
;    int            stride1,       1
;    unsigned char *src2,          2
;    int            stride2,       3
;    unsigned int  *variance,      4
;    unsigned int  *sad,           5
;)
global sym(vp8_variance_and_sad_16x16_sse2)
sym(vp8_variance_and_sad_16x16_sse2):
    push        rbp
    mov         rbp, rsp
    SHADOW_ARGS_TO_STACK 6
    GET_GOT     rbx
    push        rsi
    push        rdi
    ; end prolog

    mov         rax,        arg(0)          ; src1
    mov         rcx,        arg(1)          ; stride1
    mov         rdx,        arg(2)          ; src2
    mov         rdi,        arg(3)          ; stride2

    mov         rsi,        16              ; block height

    ; Prep accumulator registers
    pxor        xmm3, xmm3                  ; SAD
    pxor        xmm4, xmm4                  ; sum of src2
    pxor        xmm5, xmm5                  ; sum of src2^2

    ; Because we're working with the actual output frames
    ; we can't depend on any kind of data alignment.
.accumulate
    movdqa      xmm0, [rax]                 ; src1
    movdqa      xmm1, [rdx]                 ; src2
    add         rax, rcx                    ; src1 + stride1
    add         rdx, rdi                    ; src2 + stride2

    ; SAD(src1, src2)
    psadbw      xmm0, xmm1
    paddusw     xmm3, xmm0

    ; SUM(src2)
    pxor        xmm2, xmm2
    psadbw      xmm2, xmm1                  ; sum src2 by misusing SAD against 0
    paddusw     xmm4, xmm2

    ; pmaddubsw would be ideal if it took two unsigned values. instead,
    ; it expects a signed and an unsigned value. so instead we zero extend
    ; and operate on words.
    pxor        xmm2, xmm2
    movdqa      xmm0, xmm1
    punpcklbw   xmm0, xmm2
    punpckhbw   xmm1, xmm2
    pmaddwd     xmm0, xmm0
    pmaddwd     xmm1, xmm1
    paddd       xmm5, xmm0
    paddd       xmm5, xmm1

    sub         rsi,        1
    jnz         .accumulate

    ; phaddd only operates on adjacent double words.
    ; Finalize SAD and store
    movdqa      xmm0, xmm3
    psrldq      xmm0, 8
    paddusw     xmm0, xmm3
    paddd       xmm0, [GLOBAL(t128)]
    psrld       xmm0, 8

    mov         rax,  arg(5)
    movd        [rax], xmm0

    ; Accumulate sum of src2
    movdqa      xmm0, xmm4
    psrldq      xmm0, 8
    paddusw     xmm0, xmm4
    ; Square src2. Ignore high value
    pmuludq     xmm0, xmm0
    psrld       xmm0, 8

    ; phaddw could be used to sum adjacent values but we want
    ; all the values summed. promote to doubles, accumulate,
    ; shift and sum
    pxor        xmm2, xmm2
    movdqa      xmm1, xmm5
    punpckldq   xmm1, xmm2
    punpckhdq   xmm5, xmm2
    paddd       xmm1, xmm5
    movdqa      xmm2, xmm1
    psrldq      xmm1, 8
    paddd       xmm1, xmm2

    psubd       xmm1, xmm0

    ; (variance + 128) >> 8
    paddd       xmm1, [GLOBAL(t128)]
    psrld       xmm1, 8
    mov         rax,  arg(4)

    movd        [rax], xmm1


    ; begin epilog
    pop         rdi
    pop         rsi
    RESTORE_GOT
    UNSHADOW_ARGS
    pop         rbp
    ret

SECTION_RODATA
align 16
t128:
    ddq 128
align 16
tMFQE: ; 1 << MFQE_PRECISION
    times 8 dw 0x10
align 16
tMFQE_round: ; 1 << (MFQE_PRECISION - 1)
    times 8 dw 0x08