michael@0: ;
michael@0: ;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
michael@0: ;
michael@0: ;  Use of this source code is governed by a BSD-style license and patent
michael@0: ;  grant that can be found in the LICENSE file in the root of the source
michael@0: ;  tree. All contributing project authors may be found in the AUTHORS
michael@0: ;  file in the root of the source tree.
michael@0: ;
michael@0: 
michael@0: 
michael@0: %include "vpx_ports/x86_abi_support.asm"
michael@0: %include "vp8_asm_enc_offsets.asm"
michael@0: 
michael@0: 
michael@0: ; void vp8_fast_quantize_b_ssse3 | arg
michael@0: ;  (BLOCK  *b,                   |  0
michael@0: ;   BLOCKD *d)                   |  1
michael@0: ;
michael@0: 
michael@0: global sym(vp8_fast_quantize_b_ssse3) PRIVATE
michael@0: sym(vp8_fast_quantize_b_ssse3):
michael@0:     push        rbp
michael@0:     mov         rbp, rsp
michael@0:     GET_GOT     rbx
michael@0: 
michael@0: %if ABI_IS_32BIT
michael@0:     push        rdi
michael@0:     push        rsi
michael@0: %else
michael@0:   %if LIBVPX_YASM_WIN64
michael@0:     push        rdi
michael@0:     push        rsi
michael@0:   %endif
michael@0: %endif
michael@0:     ; end prolog
michael@0: 
michael@0: %if ABI_IS_32BIT
michael@0:     mov         rdi, arg(0)                 ; BLOCK *b
michael@0:     mov         rsi, arg(1)                 ; BLOCKD *d
michael@0: %else
michael@0:   %if LIBVPX_YASM_WIN64
michael@0:     mov         rdi, rcx                    ; BLOCK *b
michael@0:     mov         rsi, rdx                    ; BLOCKD *d
michael@0:   %else
michael@0:     ;mov         rdi, rdi                    ; BLOCK *b
michael@0:     ;mov         rsi, rsi                    ; BLOCKD *d
michael@0:   %endif
michael@0: %endif
michael@0: 
michael@0:     mov         rax, [rdi + vp8_block_coeff]
michael@0:     mov         rcx, [rdi + vp8_block_round]
michael@0:     mov         rdx, [rdi + vp8_block_quant_fast]
michael@0: 
michael@0:     ; coeff
michael@0:     movdqa      xmm0, [rax]
michael@0:     movdqa      xmm4, [rax + 16]
michael@0: 
michael@0:     ; round
michael@0:     movdqa      xmm2, [rcx]
michael@0:     movdqa      xmm3, [rcx + 16]
michael@0: 
michael@0:     movdqa      xmm1, xmm0
michael@0:     movdqa      xmm5, xmm4
michael@0: 
michael@0:     ; sz = z >> 15
michael@0:     psraw       xmm0, 15
michael@0:     psraw       xmm4, 15
michael@0: 
michael@0:     pabsw       xmm1, xmm1
michael@0:     pabsw       xmm5, xmm5
michael@0: 
michael@0:     paddw       xmm1, xmm2
michael@0:     paddw       xmm5, xmm3
michael@0: 
michael@0:     ; quant_fast
michael@0:     pmulhw      xmm1, [rdx]
michael@0:     pmulhw      xmm5, [rdx + 16]
michael@0: 
michael@0:     mov         rax, [rsi + vp8_blockd_qcoeff]
michael@0:     mov         rdi, [rsi + vp8_blockd_dequant]
michael@0:     mov         rcx, [rsi + vp8_blockd_dqcoeff]
michael@0: 
michael@0:     movdqa      xmm2, xmm1                  ;store y for getting eob
michael@0:     movdqa      xmm3, xmm5
michael@0: 
michael@0:     pxor        xmm1, xmm0
michael@0:     pxor        xmm5, xmm4
michael@0:     psubw       xmm1, xmm0
michael@0:     psubw       xmm5, xmm4
michael@0: 
michael@0:     movdqa      [rax], xmm1
michael@0:     movdqa      [rax + 16], xmm5
michael@0: 
michael@0:     movdqa      xmm0, [rdi]
michael@0:     movdqa      xmm4, [rdi + 16]
michael@0: 
michael@0:     pmullw      xmm0, xmm1
michael@0:     pmullw      xmm4, xmm5
michael@0:     pxor        xmm1, xmm1
michael@0: 
michael@0:     pcmpgtw     xmm2, xmm1                  ;calculate eob
michael@0:     pcmpgtw     xmm3, xmm1
michael@0:     packsswb    xmm2, xmm3
michael@0:     pshufb      xmm2, [GLOBAL(zz_shuf)]
michael@0: 
michael@0:     pmovmskb    edx, xmm2
michael@0: 
michael@0:     movdqa      [rcx], xmm0                 ;store dqcoeff
michael@0:     movdqa      [rcx + 16], xmm4            ;store dqcoeff
michael@0:     mov         rcx, [rsi + vp8_blockd_eob]
michael@0: 
michael@0:     bsr         eax, edx                    ;count 0
michael@0:     add         eax, 1
michael@0: 
michael@0:     cmp         edx, 0                      ;if all 0, eob=0
michael@0:     cmove       eax, edx
michael@0: 
michael@0:     mov         BYTE PTR [rcx], al          ;store eob
michael@0: 
michael@0:     ; begin epilog
michael@0: %if ABI_IS_32BIT
michael@0:     pop         rsi
michael@0:     pop         rdi
michael@0: %else
michael@0:   %if LIBVPX_YASM_WIN64
michael@0:     pop         rsi
michael@0:     pop         rdi
michael@0:   %endif
michael@0: %endif
michael@0: 
michael@0:     RESTORE_GOT
michael@0:     pop         rbp
michael@0:     ret
michael@0: 
michael@0: SECTION_RODATA
michael@0: align 16
michael@0: zz_shuf:
michael@0:     db 0, 1, 4, 8, 5, 2, 3, 6, 9, 12, 13, 10, 7, 11, 14, 15