generic-library/ffmpeg
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

x86inc: activate REP_RET automatically

Browse Source 
Now RET checks whether it immediately follows a branch, so the
programmer dosen't have to keep track of that condition. REP_RET
is still needed manually when it's a branch target, but that's
much rarer.

The implementation involves lots of spurious labels, but that's OK
because we strip them.

Signed-off-by: Derek Buitenhuis <derek.buitenhuis@gmail.com>
This commit is contained in:
Loren Merritt 2013-09-11 17:49:20 +02:00 committed by Derek Buitenhuis
parent ce1e8045e0
commit 25cb0c1a1e
1 changed files with 31 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

36
libavutil/x86/x86inc.asm
View File

@ -135,8 +135,7 @@ 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
@ -484,7 +483,7 @@ 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 ;=============================================
@ -531,7 +530,7 @@ 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 ;==============================================================
@ -587,7 +586,7 @@ 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 ;======================================================================
@ -601,6 +600,10 @@ 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
@ -609,6 +612,29 @@ 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