virtualenv/FFmpeg
1
0
Fork 0
mirror of https://github.com/FFmpeg/FFmpeg.git synced 2025-08-10 06:10:52 +02:00
Code Issues Releases Activity

Merge commit '25cb0c1a1e66edacc1667acf6818f524c0997f10'

Browse Source 
* commit '25cb0c1a1e66edacc1667acf6818f524c0997f10':
  x86inc: activate REP_RET automatically

Merged-by: Michael Niedermayer <michaelni@gmx.at>
This commit is contained in:
Michael Niedermayer
2013-10-08 10:27:20 +02:00
parent ba9c557b92 25cb0c1a1e
commit c1488fab3d
1 changed files with 31 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
libavutil/x86/x86inc.asm
View File

@@ -154,8 +154,7 @@ CPUNOP amdnop
; Pops anything that was pushed by PROLOGUE, and returns. ; Pops anything that was pushed by PROLOGUE, and returns.
; REP_RET: ; REP_RET:
; Same, but if it doesn't pop anything it becomes a 2-byte ret, for athlons ; Use this instead of RET if it's a branch target.
; which are slow when a normal ret follows a branch.
; registers: ; registers:
; rN and rNq are the native-size register holding function argument N ; rN and rNq are the native-size register holding function argument N
@@ -503,7 +502,7 @@ DECLARE_REG 14, R15, 120
%if mmsize == 32 %if mmsize == 32
vzeroupper vzeroupper
%endif %endif
ret AUTO_REP_RET
%endmacro %endmacro
%elif ARCH_X86_64 ; *nix x64 ;============================================= %elif ARCH_X86_64 ; *nix x64 ;=============================================
@@ -550,7 +549,7 @@ DECLARE_REG 14, R15, 72
%if mmsize == 32 %if mmsize == 32
vzeroupper vzeroupper
%endif %endif
ret AUTO_REP_RET
%endmacro %endmacro
%else ; X86_32 ;============================================================== %else ; X86_32 ;==============================================================
@@ -606,7 +605,7 @@ DECLARE_ARG 7, 8, 9, 10, 11, 12, 13, 14
%if mmsize == 32 %if mmsize == 32
vzeroupper vzeroupper
%endif %endif
ret AUTO_REP_RET
%endmacro %endmacro
%endif ;====================================================================== %endif ;======================================================================
@@ -620,6 +619,10 @@ DECLARE_ARG 7, 8, 9, 10, 11, 12, 13, 14
%endmacro %endmacro
%endif %endif
; On AMD cpus <=K10, an ordinary ret is slow if it immediately follows either
; a branch or a branch target. So switch to a 2-byte form of ret in that case.
; We can automatically detect "follows a branch", but not a branch target.
; (SSSE3 is a sufficient condition to know that your cpu doesn't have this problem.)
%macro REP_RET 0 %macro REP_RET 0
%if has_epilogue %if has_epilogue
RET RET
@@ -628,6 +631,29 @@ DECLARE_ARG 7, 8, 9, 10, 11, 12, 13, 14
%endif %endif
%endmacro %endmacro
%define last_branch_adr $$
%macro AUTO_REP_RET 0
%ifndef cpuflags
times ((last_branch_adr-$)>>31)+1 rep ; times 1 iff $ != last_branch_adr.
%elif notcpuflag(ssse3)
times ((last_branch_adr-$)>>31)+1 rep
%endif
ret
%endmacro
%macro BRANCH_INSTR 0-*
%rep %0
%macro %1 1-2 %1
%2 %1
%%branch_instr:
%xdefine last_branch_adr %%branch_instr
%endmacro
%rotate 1
%endrep
%endmacro
BRANCH_INSTR jz, je, jnz, jne, jl, jle, jnl, jnle, jg, jge, jng, jnge, ja, jae, jna, jnae, jb, jbe, jnb, jnbe, jc, jnc, js, jns, jo, jno, jp, jnp
%macro TAIL_CALL 2 ; callee, is_nonadjacent %macro TAIL_CALL 2 ; callee, is_nonadjacent
%if has_epilogue %if has_epilogue
call %1 call %1