1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libvpx/vp8/encoder/x86/quantize_ssse3.asm Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,138 @@
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_fast_quantize_b_ssse3 | arg
1.19 +; (BLOCK *b, | 0
1.20 +; BLOCKD *d) | 1
1.21 +;
1.22 +
1.23 +global sym(vp8_fast_quantize_b_ssse3) PRIVATE
1.24 +sym(vp8_fast_quantize_b_ssse3):
1.25 + push rbp
1.26 + mov rbp, rsp
1.27 + GET_GOT rbx
1.28 +
1.29 +%if ABI_IS_32BIT
1.30 + push rdi
1.31 + push rsi
1.32 +%else
1.33 + %if LIBVPX_YASM_WIN64
1.34 + push rdi
1.35 + push rsi
1.36 + %endif
1.37 +%endif
1.38 + ; end prolog
1.39 +
1.40 +%if ABI_IS_32BIT
1.41 + mov rdi, arg(0) ; BLOCK *b
1.42 + mov rsi, arg(1) ; BLOCKD *d
1.43 +%else
1.44 + %if LIBVPX_YASM_WIN64
1.45 + mov rdi, rcx ; BLOCK *b
1.46 + mov rsi, rdx ; BLOCKD *d
1.47 + %else
1.48 + ;mov rdi, rdi ; BLOCK *b
1.49 + ;mov rsi, rsi ; BLOCKD *d
1.50 + %endif
1.51 +%endif
1.52 +
1.53 + mov rax, [rdi + vp8_block_coeff]
1.54 + mov rcx, [rdi + vp8_block_round]
1.55 + mov rdx, [rdi + vp8_block_quant_fast]
1.56 +
1.57 + ; coeff
1.58 + movdqa xmm0, [rax]
1.59 + movdqa xmm4, [rax + 16]
1.60 +
1.61 + ; round
1.62 + movdqa xmm2, [rcx]
1.63 + movdqa xmm3, [rcx + 16]
1.64 +
1.65 + movdqa xmm1, xmm0
1.66 + movdqa xmm5, xmm4
1.67 +
1.68 + ; sz = z >> 15
1.69 + psraw xmm0, 15
1.70 + psraw xmm4, 15
1.71 +
1.72 + pabsw xmm1, xmm1
1.73 + pabsw xmm5, xmm5
1.74 +
1.75 + paddw xmm1, xmm2
1.76 + paddw xmm5, xmm3
1.77 +
1.78 + ; quant_fast
1.79 + pmulhw xmm1, [rdx]
1.80 + pmulhw xmm5, [rdx + 16]
1.81 +
1.82 + mov rax, [rsi + vp8_blockd_qcoeff]
1.83 + mov rdi, [rsi + vp8_blockd_dequant]
1.84 + mov rcx, [rsi + vp8_blockd_dqcoeff]
1.85 +
1.86 + movdqa xmm2, xmm1 ;store y for getting eob
1.87 + movdqa xmm3, xmm5
1.88 +
1.89 + pxor xmm1, xmm0
1.90 + pxor xmm5, xmm4
1.91 + psubw xmm1, xmm0
1.92 + psubw xmm5, xmm4
1.93 +
1.94 + movdqa [rax], xmm1
1.95 + movdqa [rax + 16], xmm5
1.96 +
1.97 + movdqa xmm0, [rdi]
1.98 + movdqa xmm4, [rdi + 16]
1.99 +
1.100 + pmullw xmm0, xmm1
1.101 + pmullw xmm4, xmm5
1.102 + pxor xmm1, xmm1
1.103 +
1.104 + pcmpgtw xmm2, xmm1 ;calculate eob
1.105 + pcmpgtw xmm3, xmm1
1.106 + packsswb xmm2, xmm3
1.107 + pshufb xmm2, [GLOBAL(zz_shuf)]
1.108 +
1.109 + pmovmskb edx, xmm2
1.110 +
1.111 + movdqa [rcx], xmm0 ;store dqcoeff
1.112 + movdqa [rcx + 16], xmm4 ;store dqcoeff
1.113 + mov rcx, [rsi + vp8_blockd_eob]
1.114 +
1.115 + bsr eax, edx ;count 0
1.116 + add eax, 1
1.117 +
1.118 + cmp edx, 0 ;if all 0, eob=0
1.119 + cmove eax, edx
1.120 +
1.121 + mov BYTE PTR [rcx], al ;store eob
1.122 +
1.123 + ; begin epilog
1.124 +%if ABI_IS_32BIT
1.125 + pop rsi
1.126 + pop rdi
1.127 +%else
1.128 + %if LIBVPX_YASM_WIN64
1.129 + pop rsi
1.130 + pop rdi
1.131 + %endif
1.132 +%endif
1.133 +
1.134 + RESTORE_GOT
1.135 + pop rbp
1.136 + ret
1.137 +
1.138 +SECTION_RODATA
1.139 +align 16
1.140 +zz_shuf:
1.141 + db 0, 1, 4, 8, 5, 2, 3, 6, 9, 12, 13, 10, 7, 11, 14, 15
