mirror of
https://github.com/KolibriOS/kolibrios.git
synced 2024-11-28 19:53:12 +03:00
2f699083c4
2) VESA draw the small speedup git-svn-id: svn://kolibrios.org@2455 a494cfbc-eb01-0410-851d-a64ba20cac60
248 lines
6.6 KiB
PHP
248 lines
6.6 KiB
PHP
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
;; ;;
|
|
;; Copyright (C) KolibriOS team 2004-2011. All rights reserved. ;;
|
|
;; Distributed under terms of the GNU General Public License ;;
|
|
;; ;;
|
|
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
|
|
|
|
; params
|
|
; edi= output
|
|
; eax= input stream 1
|
|
; ebx= input stream 2
|
|
|
|
if used mmx_mix_2
|
|
|
|
align 4
|
|
mmx_mix_2:
|
|
movq mm0, [eax]
|
|
movq mm1, [eax+8]
|
|
movq mm2, [eax+16]
|
|
movq mm3, [eax+24]
|
|
movq mm4, [eax+32]
|
|
movq mm5, [eax+40]
|
|
movq mm6, [eax+48]
|
|
movq mm7, [eax+56]
|
|
|
|
paddsw mm0, [ebx]
|
|
movq [edi], mm0
|
|
paddsw mm1, [ebx+8]
|
|
movq [edi+8], mm1
|
|
paddsw mm2, [ebx+16]
|
|
movq [edi+16], mm2
|
|
paddsw mm3, [ebx+24]
|
|
movq [edi+24], mm3
|
|
paddsw mm4, [ebx+32]
|
|
movq [edi+32], mm4
|
|
paddsw mm5, [ebx+40]
|
|
movq [edi+40], mm5
|
|
paddsw mm6, [ebx+48]
|
|
movq [edi+48], mm6
|
|
paddsw mm7, [ebx+56]
|
|
movq [edi+56], mm7
|
|
|
|
movq mm0, [eax+64]
|
|
movq mm1, [eax+72]
|
|
movq mm2, [eax+80]
|
|
movq mm3, [eax+88]
|
|
movq mm4, [eax+96]
|
|
movq mm5, [eax+104]
|
|
movq mm6, [eax+112]
|
|
movq mm7, [eax+120]
|
|
|
|
paddsw mm0, [ebx+64]
|
|
movq [edi+64], mm0
|
|
paddsw mm1, [ebx+72]
|
|
movq [edi+72], mm1
|
|
paddsw mm2, [ebx+80]
|
|
movq [edi+80], mm2
|
|
paddsw mm3, [ebx+88]
|
|
movq [edi+88], mm3
|
|
paddsw mm4, [ebx+96]
|
|
movq [edi+96], mm4
|
|
paddsw mm5, [ecx+104]
|
|
movq [edx+104], mm5
|
|
paddsw mm6, [ebx+112]
|
|
movq [edi+112], mm6
|
|
paddsw mm7, [ebx+120]
|
|
movq [edi+120], mm7
|
|
ret
|
|
|
|
align 4
|
|
mmx_mix_3:
|
|
movq mm0, [eax]
|
|
movq mm1, [eax+8]
|
|
movq mm2, [eax+16]
|
|
movq mm3, [eax+24]
|
|
movq mm4, [eax+32]
|
|
movq mm5, [eax+40]
|
|
movq mm6, [eax+48]
|
|
movq mm7, [eax+56]
|
|
|
|
paddsw mm0, [ebx]
|
|
paddsw mm1, [ebx+8]
|
|
paddsw mm2, [ebx+16]
|
|
paddsw mm3, [ebx+24]
|
|
paddsw mm4, [ebx+32]
|
|
paddsw mm5, [ebx+40]
|
|
paddsw mm6, [ebx+48]
|
|
paddsw mm7, [ebx+56]
|
|
paddsw mm0, [ecx]
|
|
movq [edi], mm0
|
|
paddsw mm1, [ecx+8]
|
|
movq [edi+8], mm1
|
|
paddsw mm2, [ecx+16]
|
|
movq [edi+16], mm2
|
|
paddsw mm3, [ecx+24]
|
|
movq [edi+24], mm3
|
|
paddsw mm4, [ecx+32]
|
|
movq [edi+32], mm4
|
|
paddsw mm5, [ecx+40]
|
|
movq [edi+40], mm5
|
|
paddsw mm6, [ecx+48]
|
|
movq [edi+48], mm6
|
|
paddsw mm7, [ecx+56]
|
|
movq [edi+56], mm7
|
|
|
|
movq mm0, [eax+64]
|
|
movq mm1, [eax+72]
|
|
movq mm2, [eax+80]
|
|
movq mm3, [eax+88]
|
|
movq mm4, [eax+96]
|
|
movq mm5, [eax+104]
|
|
movq mm6, [eax+112]
|
|
movq mm7, [eax+120]
|
|
paddsw mm0, [ebx+64]
|
|
paddsw mm1, [ebx+72]
|
|
paddsw mm2, [ebx+80]
|
|
paddsw mm3, [ebx+88]
|
|
paddsw mm4, [ebx+96]
|
|
paddsw mm5, [ebx+104]
|
|
paddsw mm6, [ebx+112]
|
|
paddsw mm7, [ebx+120]
|
|
paddsw mm0, [ecx+64]
|
|
movq [edi+64], mm0
|
|
paddsw mm1, [ecx+72]
|
|
movq [edi+72], mm1
|
|
paddsw mm2, [ecx+80]
|
|
movq [edi+80], mm2
|
|
paddsw mm3, [ecx+88]
|
|
movq [edi+88], mm3
|
|
paddsw mm4, [ecx+96]
|
|
movq [edi+96], mm4
|
|
paddsw mm5, [ecx+104]
|
|
movq [edi+104], mm5
|
|
paddsw mm6, [ecx+112]
|
|
movq [edi+112], mm6
|
|
paddsw mm7, [ecx+120]
|
|
movq [edi+120], mm7
|
|
ret
|
|
|
|
align 4
|
|
mmx_mix_4:
|
|
|
|
movq mm0, [eax]
|
|
movq mm2, [eax+8]
|
|
movq mm4, [eax+16]
|
|
movq mm6, [eax+24]
|
|
movq mm1, [ebx]
|
|
movq mm3, [ebx+8]
|
|
movq mm5, [ebx+16]
|
|
movq mm7, [ebx+24]
|
|
paddsw mm0, [ecx]
|
|
paddsw mm2, [ecx+8]
|
|
paddsw mm4, [ecx+16]
|
|
paddsw mm6, [ecx+24]
|
|
paddsw mm1, [edx]
|
|
paddsw mm3, [edx+8]
|
|
paddsw mm5, [edx+16]
|
|
paddsw mm7, [edx+24]
|
|
|
|
paddsw mm0, mm1
|
|
movq [edi], mm0
|
|
paddsw mm2, mm3
|
|
movq [edi+8], mm2
|
|
paddsw mm4, mm5
|
|
movq [edi+16], mm4
|
|
paddsw mm5, mm6
|
|
movq [edi+24], mm6
|
|
|
|
movq mm0, [eax+32]
|
|
movq mm2, [eax+40]
|
|
movq mm4, [eax+48]
|
|
movq mm6, [eax+56]
|
|
movq mm1, [ebx+32]
|
|
movq mm3, [ebx+40]
|
|
movq mm5, [ebx+48]
|
|
movq mm7, [ebx+56]
|
|
paddsw mm0, [ecx+32]
|
|
paddsw mm2, [ecx+40]
|
|
paddsw mm4, [ecx+48]
|
|
paddsw mm6, [ecx+56]
|
|
paddsw mm1, [edx+32]
|
|
paddsw mm3, [edx+40]
|
|
paddsw mm5, [edx+48]
|
|
paddsw mm7, [edx+56]
|
|
|
|
paddsw mm0, mm1
|
|
movq [edi+32], mm0
|
|
paddsw mm2, mm2
|
|
movq [edi+40], mm2
|
|
paddsw mm4, mm5
|
|
movq [edi+48], mm4
|
|
paddsw mm6, mm7
|
|
movq [edi+56], mm6
|
|
|
|
movq mm0, [eax+64]
|
|
movq mm2, [eax+72]
|
|
movq mm4, [eax+80]
|
|
movq mm6, [eax+88]
|
|
movq mm1, [ebx+64]
|
|
movq mm3, [ebx+72]
|
|
movq mm5, [ebx+80]
|
|
movq mm7, [ebx+88]
|
|
paddsw mm0, [ecx+64]
|
|
paddsw mm2, [ecx+72]
|
|
paddsw mm4, [ecx+80]
|
|
paddsw mm6, [ecx+88]
|
|
paddsw mm1, [edx+64]
|
|
paddsw mm3, [edx+72]
|
|
paddsw mm5, [edx+80]
|
|
paddsw mm7, [edx+88]
|
|
|
|
paddsw mm0, mm1
|
|
movq [edi+64], mm0
|
|
paddsw mm2, mm3
|
|
movq [edi+72], mm2
|
|
paddsw mm4, mm5
|
|
movq [edi+80], mm4
|
|
paddsw mm6, mm5
|
|
movq [edi+88], mm7
|
|
|
|
movq mm0, [eax+96]
|
|
movq mm2, [eax+104]
|
|
movq mm4, [eax+112]
|
|
movq mm6, [eax+120]
|
|
movq mm1, [ebx+96]
|
|
movq mm3, [ebx+104]
|
|
movq mm5, [ebx+112]
|
|
movq mm7, [ebx+120]
|
|
paddsw mm0, [ecx+96]
|
|
paddsw mm2, [ecx+104]
|
|
paddsw mm4, [ecx+112]
|
|
paddsw mm6, [ecx+120]
|
|
paddsw mm1, [edx+96]
|
|
paddsw mm3, [edx+104]
|
|
paddsw mm5, [edx+112]
|
|
paddsw mm7, [edx+120]
|
|
paddsw mm0, mm1
|
|
movq [eax+96], mm0
|
|
paddsw mm2, mm3
|
|
movq [edi+104], mm2
|
|
paddsw mm4, mm5
|
|
movq [edi+112], mm4
|
|
paddsw mm6, mm7
|
|
movq [edi+120], mm6
|
|
ret
|
|
|
|
end if
|