1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp8/encoder/x86/quantize_sse4.asm Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,256 @@
1.4 +;
1.5 +; Copyright (c) 2010 The WebM project authors. All Rights Reserved.
1.6 +;
1.7 +; Use of this source code is governed by a BSD-style license and patent
1.8 +; grant that can be found in the LICENSE file in the root of the source
1.9 +; tree. All contributing project authors may be found in the AUTHORS
1.10 +; file in the root of the source tree.
1.11 +;
1.12 +
1.13 +
1.14 +%include "vpx_ports/x86_abi_support.asm"
1.15 +%include "vp8_asm_enc_offsets.asm"
1.16 +
1.17 +
1.18 +; void vp8_regular_quantize_b_sse4 | arg
1.19 +; (BLOCK *b, | 0
1.20 +; BLOCKD *d) | 1
1.21 +
1.22 +global sym(vp8_regular_quantize_b_sse4) PRIVATE
1.23 +sym(vp8_regular_quantize_b_sse4):
1.24 +
1.25 +%if ABI_IS_32BIT
1.26 + push rbp
1.27 + mov rbp, rsp
1.28 + GET_GOT rbx
1.29 + push rdi
1.30 + push rsi
1.31 +
1.32 + ALIGN_STACK 16, rax
1.33 + %define qcoeff 0 ; 32
1.34 + %define stack_size 32
1.35 + sub rsp, stack_size
1.36 +%else
1.37 + %if LIBVPX_YASM_WIN64
1.38 + SAVE_XMM 8, u
1.39 + push rdi
1.40 + push rsi
1.41 + %endif
1.42 +%endif
1.43 + ; end prolog
1.44 +
1.45 +%if ABI_IS_32BIT
1.46 + mov rdi, arg(0) ; BLOCK *b
1.47 + mov rsi, arg(1) ; BLOCKD *d
1.48 +%else
1.49 + %if LIBVPX_YASM_WIN64
1.50 + mov rdi, rcx ; BLOCK *b
1.51 + mov rsi, rdx ; BLOCKD *d
1.52 + %else
1.53 + ;mov rdi, rdi ; BLOCK *b
1.54 + ;mov rsi, rsi ; BLOCKD *d
1.55 + %endif
1.56 +%endif
1.57 +
1.58 + mov rax, [rdi + vp8_block_coeff]
1.59 + mov rcx, [rdi + vp8_block_zbin]
1.60 + mov rdx, [rdi + vp8_block_round]
1.61 + movd xmm7, [rdi + vp8_block_zbin_extra]
1.62 +
1.63 + ; z
1.64 + movdqa xmm0, [rax]
1.65 + movdqa xmm1, [rax + 16]
1.66 +
1.67 + ; duplicate zbin_oq_value
1.68 + pshuflw xmm7, xmm7, 0
1.69 + punpcklwd xmm7, xmm7
1.70 +
1.71 + movdqa xmm2, xmm0
1.72 + movdqa xmm3, xmm1
1.73 +
1.74 + ; sz
1.75 + psraw xmm0, 15
1.76 + psraw xmm1, 15
1.77 +
1.78 + ; (z ^ sz)
1.79 + pxor xmm2, xmm0
1.80 + pxor xmm3, xmm1
1.81 +
1.82 + ; x = abs(z)
1.83 + psubw xmm2, xmm0
1.84 + psubw xmm3, xmm1
1.85 +
1.86 + ; zbin
1.87 + movdqa xmm4, [rcx]
1.88 + movdqa xmm5, [rcx + 16]
1.89 +
1.90 + ; *zbin_ptr + zbin_oq_value
1.91 + paddw xmm4, xmm7
1.92 + paddw xmm5, xmm7
1.93 +
1.94 + movdqa xmm6, xmm2
1.95 + movdqa xmm7, xmm3
1.96 +
1.97 + ; x - (*zbin_ptr + zbin_oq_value)
1.98 + psubw xmm6, xmm4
1.99 + psubw xmm7, xmm5
1.100 +
1.101 + ; round
1.102 + movdqa xmm4, [rdx]
1.103 + movdqa xmm5, [rdx + 16]
1.104 +
1.105 + mov rax, [rdi + vp8_block_quant_shift]
1.106 + mov rcx, [rdi + vp8_block_quant]
1.107 + mov rdx, [rdi + vp8_block_zrun_zbin_boost]
1.108 +
1.109 + ; x + round
1.110 + paddw xmm2, xmm4
1.111 + paddw xmm3, xmm5
1.112 +
1.113 + ; quant
1.114 + movdqa xmm4, [rcx]
1.115 + movdqa xmm5, [rcx + 16]
1.116 +
1.117 + ; y = x * quant_ptr >> 16
1.118 + pmulhw xmm4, xmm2
1.119 + pmulhw xmm5, xmm3
1.120 +
1.121 + ; y += x
1.122 + paddw xmm2, xmm4
1.123 + paddw xmm3, xmm5
1.124 +
1.125 + pxor xmm4, xmm4
1.126 +%if ABI_IS_32BIT
1.127 + movdqa [rsp + qcoeff], xmm4
1.128 + movdqa [rsp + qcoeff + 16], xmm4
1.129 +%else
1.130 + pxor xmm8, xmm8
1.131 +%endif
1.132 +
1.133 + ; quant_shift
1.134 + movdqa xmm5, [rax]
1.135 +
1.136 + ; zrun_zbin_boost
1.137 + mov rax, rdx
1.138 +
1.139 +%macro ZIGZAG_LOOP 5
1.140 + ; x
1.141 + pextrw ecx, %4, %2
1.142 +
1.143 + ; if (x >= zbin)
1.144 + sub cx, WORD PTR[rdx] ; x - zbin
1.145 + lea rdx, [rdx + 2] ; zbin_boost_ptr++
1.146 + jl .rq_zigzag_loop_%1 ; x < zbin
1.147 +
1.148 + pextrw edi, %3, %2 ; y
1.149 +
1.150 + ; downshift by quant_shift[rc]
1.151 + pextrb ecx, xmm5, %1 ; quant_shift[rc]
1.152 + sar edi, cl ; also sets Z bit
1.153 + je .rq_zigzag_loop_%1 ; !y
1.154 +%if ABI_IS_32BIT
1.155 + mov WORD PTR[rsp + qcoeff + %1 *2], di
1.156 +%else
1.157 + pinsrw %5, edi, %2 ; qcoeff[rc]
1.158 +%endif
1.159 + mov rdx, rax ; reset to b->zrun_zbin_boost
1.160 +.rq_zigzag_loop_%1:
1.161 +%endmacro
1.162 +; in vp8_default_zig_zag1d order: see vp8/common/entropy.c
1.163 +ZIGZAG_LOOP 0, 0, xmm2, xmm6, xmm4
1.164 +ZIGZAG_LOOP 1, 1, xmm2, xmm6, xmm4
1.165 +ZIGZAG_LOOP 4, 4, xmm2, xmm6, xmm4
1.166 +ZIGZAG_LOOP 8, 0, xmm3, xmm7, xmm8
1.167 +ZIGZAG_LOOP 5, 5, xmm2, xmm6, xmm4
1.168 +ZIGZAG_LOOP 2, 2, xmm2, xmm6, xmm4
1.169 +ZIGZAG_LOOP 3, 3, xmm2, xmm6, xmm4
1.170 +ZIGZAG_LOOP 6, 6, xmm2, xmm6, xmm4
1.171 +ZIGZAG_LOOP 9, 1, xmm3, xmm7, xmm8
1.172 +ZIGZAG_LOOP 12, 4, xmm3, xmm7, xmm8
1.173 +ZIGZAG_LOOP 13, 5, xmm3, xmm7, xmm8
1.174 +ZIGZAG_LOOP 10, 2, xmm3, xmm7, xmm8
1.175 +ZIGZAG_LOOP 7, 7, xmm2, xmm6, xmm4
1.176 +ZIGZAG_LOOP 11, 3, xmm3, xmm7, xmm8
1.177 +ZIGZAG_LOOP 14, 6, xmm3, xmm7, xmm8
1.178 +ZIGZAG_LOOP 15, 7, xmm3, xmm7, xmm8
1.179 +
1.180 + mov rcx, [rsi + vp8_blockd_dequant]
1.181 + mov rdi, [rsi + vp8_blockd_dqcoeff]
1.182 +
1.183 +%if ABI_IS_32BIT
1.184 + movdqa xmm4, [rsp + qcoeff]
1.185 + movdqa xmm5, [rsp + qcoeff + 16]
1.186 +%else
1.187 + %define xmm5 xmm8
1.188 +%endif
1.189 +
1.190 + ; y ^ sz
1.191 + pxor xmm4, xmm0
1.192 + pxor xmm5, xmm1
1.193 + ; x = (y ^ sz) - sz
1.194 + psubw xmm4, xmm0
1.195 + psubw xmm5, xmm1
1.196 +
1.197 + ; dequant
1.198 + movdqa xmm0, [rcx]
1.199 + movdqa xmm1, [rcx + 16]
1.200 +
1.201 + mov rcx, [rsi + vp8_blockd_qcoeff]
1.202 +
1.203 + pmullw xmm0, xmm4
1.204 + pmullw xmm1, xmm5
1.205 +
1.206 + ; store qcoeff
1.207 + movdqa [rcx], xmm4
1.208 + movdqa [rcx + 16], xmm5
1.209 +
1.210 + ; store dqcoeff
1.211 + movdqa [rdi], xmm0
1.212 + movdqa [rdi + 16], xmm1
1.213 +
1.214 + mov rcx, [rsi + vp8_blockd_eob]
1.215 +
1.216 + ; select the last value (in zig_zag order) for EOB
1.217 + pxor xmm6, xmm6
1.218 + pcmpeqw xmm4, xmm6
1.219 + pcmpeqw xmm5, xmm6
1.220 +
1.221 + packsswb xmm4, xmm5
1.222 + pshufb xmm4, [GLOBAL(zig_zag1d)]
1.223 + pmovmskb edx, xmm4
1.224 + xor rdi, rdi
1.225 + mov eax, -1
1.226 + xor dx, ax
1.227 + bsr eax, edx
1.228 + sub edi, edx
1.229 + sar edi, 31
1.230 + add eax, 1
1.231 + and eax, edi
1.232 +
1.233 + mov BYTE PTR [rcx], al ; store eob
1.234 +
1.235 + ; begin epilog
1.236 +%if ABI_IS_32BIT
1.237 + add rsp, stack_size
1.238 + pop rsp
1.239 +
1.240 + pop rsi
1.241 + pop rdi
1.242 + RESTORE_GOT
1.243 + pop rbp
1.244 +%else
1.245 + %undef xmm5
1.246 + %if LIBVPX_YASM_WIN64
1.247 + pop rsi
1.248 + pop rdi
1.249 + RESTORE_XMM
1.250 + %endif
1.251 +%endif
1.252 +
1.253 + ret
1.254 +
1.255 +SECTION_RODATA
1.256 +align 16
1.257 +; vp8/common/entropy.c: vp8_default_zig_zag1d
1.258 +zig_zag1d:
1.259 + db 0, 1, 4, 8, 5, 2, 3, 6, 9, 12, 13, 10, 7, 11, 14, 15
