Commit Graph

217 Commits

Author SHA1 Message Date
Licai Guo
2f8c539e60 Merge pull request from dongzha/FixIssueMcNEON
Fix potential issue for neon implement on encoder mode decision.
2014-04-17 17:26:25 +08:00
Licai Guo
4062fa9d34 Merge pull request from zhilwang/pf-test
Move copy_mb neon code to common folder
2014-04-17 11:08:56 +08:00
Licai Guo
3d9d00b27c Update targets.mk 2014-04-17 10:43:10 +08:00
Licai Guo
c8e1a41c29 Move copy_mb neon code to common folder 2014-04-17 10:06:48 +08:00
ruil2
b553468ad3 keep the declaration and definition in the same namespace 2014-04-17 09:45:26 +08:00
huili2
4ab8c88e98 divide copy_mb functions into new file for decoder use from encoder and add files for EC in decoder only. 2014-04-14 20:17:41 -07:00
Dong Zhang
8a4300be50 Fix potential issue for neon implement on encoder mode decision.
Error happens when ME_REFINE_BUF_STRIDE is not equal to 32.
2014-04-13 19:41:29 -07:00
Martin Storsjö
b35c21201b Use the Windows Runtime ThreadPool API for creating threads on Windows Phone
Windows Phone lacks the old CreateThread/beginthreadex APIs for
creating threads. (Technically, the functions still do exist,
but they aren't officially supported and aren't visible in the
headers when targeting Windows Phone.)

Building code that uses the Windows Runtime language extensions
requires building with the -ZW option.
2014-04-01 11:18:49 +03:00
Martin Storsjö
f293d26a62 Use more modern versions of functions that don't exist on Windows Phone 2014-04-01 11:18:48 +03:00
Martin Storsjö
4bcb03c5a0 Remove the unused function WelsSleep
Windows Phone 8 doesn't have Sleep(), but there's no need to
use the function at all.
2014-04-01 11:18:48 +03:00
volvet
9f50e0c91e clean multi-threading macro 2014-03-31 18:24:10 -07:00
ruil2
6b3f89d582 move some common functions to common.cpp and add some functions in common 2014-03-25 15:35:55 +08:00
Licai Guo
e39de8d404 reoranize common to inc/src/x86/arm 2014-03-18 19:41:32 -07:00
volvet
7313ecdbd0 Merge pull request from mstorsjo/use-apple-builtin-define
Use __APPLE__ instead of APPLE_IOS for apple/arm specific features
2014-03-19 09:45:56 +08:00
Licai Guo
d897d362ab Merge pull request from huili2/WELS_CLIP1
Modify MACRO WELS_CLIP1 as inline functions
2014-03-19 08:50:04 +08:00
Martin Storsjö
9586c59b9e Use __APPLE__ instead of APPLE_IOS in the arm assembly sources 2014-03-18 23:15:49 +02:00
Martin Storsjö
73ed237d73 Use __APPLE__ instead of APPLE_IOS for using the apple cpu feature detection 2014-03-18 23:15:49 +02:00
Ethan Hugg
197423f271 Merge pull request from ylatuya/master
Fix compiler warnings and remove dead code
2014-03-18 13:28:02 -07:00
Andoni Morales Alastruey
703c69de81 codec: add a new macro for unused functions
Variables used only for tracing logs can trigger
-Werror=unusef-variable when tracing is disabled.
This macro helps to silent gcc in those casesWIP
2014-03-18 19:15:25 +01:00
Martin Storsjö
e1b5e038d2 Use .obj as suffix for object files on MSVC
This avoids warnings when linking about "unrecognized source file
type, object file assumed".
2014-03-18 19:41:06 +02:00
huili2
090e8cc1ed modify WELS_CLIP1 to be inline functions 2014-03-18 01:54:25 -07:00
volvet
b21411ad7c Merge pull request from mstorsjo/remove-unused-define
Remove the unused FORMAT_COFF define
2014-03-18 16:11:22 +08:00
volvet
fb1958ad13 Merge pull request from mstorsjo/push-xmm-registers
Backup/restore the xmm6-xmm15 SSE registers within asm functions on win64

Reviewed by zhiliang
2014-03-18 15:04:54 +08:00
volvet
b5353c8455 Merge pull request from mstorsjo/fix-yasm-64bit
Fix building with yasm in 64 bit mode
2014-03-18 09:29:42 +08:00
volvet
e75cd2298b Merge pull request from mstorsjo/simplify-x86-asm-func-macro
Fold ALIGN 16 and the function label into WELS_EXTERN
2014-03-18 09:29:17 +08:00
Martin Storsjö
4633626d69 Remove XMMREG_PROTECT
This isn't necessary any longer, when all the assembly routines
take care of restoring registers as necessary.
2014-03-17 13:47:01 +02:00
Martin Storsjö
3cf52554f7 Backup/restore the xmm6-xmm15 SSE registers within asm functions on win64
According to the Win64 ABI, these registers need to be preserved,
and compilers are allowed to rely on their content to stay
available - not only for float usage but for any usage, anywhere,
in the calling C++ code.

This adds a macro which pushes the clobbered registers onto the
stack if targeting win64 (and a matching one which restores them).
The parameter to the macro is the number of xmm registers used
(e.g. if using xmm0 - xmm7, the parameter is 8), or in other
words, the number of the highest xmm register used plus one.

This is similar to how the same issue is handled for the NEON
registers q4-q7 with the vpush instruction, except that they needed
to be preserved on all platforms, not only on one particular platform.

This allows removing the XMMREG_PROTECT_* hacks, which can
easily fail if the compiler chooses to use the callee saved
xmm registers in an unexpected spot.
2014-03-17 13:44:33 +02:00
Martin Storsjö
9293f2f947 Remove commented out rodata sections and tables in assembly files 2014-03-17 13:42:18 +02:00
Martin Storsjö
eec968234d Fold ALIGN 16 and the function label into WELS_EXTERN
This simplifies the structure for all x86 assembly functions,
reducing the amount of duplicated code structure.
2014-03-17 13:35:00 +02:00
Martin Storsjö
04f5bcd68d Use movsxd in SIGN_EXTENSION
This is what nasm ended up assembling movsx with 32 bit input to
anyway.

Keep using plain movsx for 16 bit input.

This fixes building with yasm in 64 bit mode.
2014-03-17 13:26:46 +02:00
Martin Storsjö
f96918283f Remove commented out code for old, 32-bit only x86 assembly function prologues/epilogues 2014-03-17 11:20:11 +02:00
Licai Guo
b5a4d706b9 Merge pull request from mstorsjo/use-sign-extend-macro
Use the SIGN_EXTENSION macro where possible
2014-03-17 16:31:03 +08:00
Licai Guo
1c0ba88b0e Merge pull request from mstorsjo/neon-register-backup
Avoid clobbering the registers q4-q7 in DeblockingBSCalcEnc_neon
2014-03-17 14:05:23 +08:00
Martin Storsjö
fc260b39e0 Remove the unused FORMAT_COFF define
Nothing in the project currently sets FORMAT_COFF - the other generic
branch works just fine on windows.
2014-03-16 17:54:55 +02:00
Martin Storsjö
eb238e6549 Use the SIGN_EXTENSION macro where possible
This shortens the x86 assembly by 134 lines in total.
2014-03-16 17:54:24 +02:00
Martin Storsjö
91e5838621 Indent all WELS_ASM_FUNC_BEGIN properly
By having all of them start at the start of the line, the code
is more consistent and readable.
2014-03-16 12:01:54 +02:00
Martin Storsjö
c82f548e6f Add defines of arg11 and arg12 in asm_inc.asm 2014-03-15 14:42:07 +02:00
Martin Storsjö
cde30c155b Avoid clobbering the registers q4-q7 in DeblockingBSCalcEnc_neon
Remap q5 to q8, q6 to q9, q7 to q10 and q8 to q11, and push
q4 to the stack.

This was missed previously since the codec unittest doesn't
test encoding with loop filter enabled yet.
2014-03-14 22:22:28 +02:00
Martin Storsjö
9199798f22 Fix a typo in a macro name, EXTENTION -> EXTENSION 2014-03-14 10:13:18 +02:00
volvet
6714b8ae99 Merge pull request from mstorsjo/dont-clobber-neon-registers
Avoid clobbering the neon registers q4-q7

Review and verified by zhilwang
2014-03-14 10:28:55 +08:00
Martin Storsjö
efe32b7900 Make arm assembly labels always start from the beginning of the line
A few labels were misformatted.
2014-03-12 12:01:01 +02:00
Martin Storsjö
52e8973869 Mark the stack as non-executable in the arm assembly
Otherwise the linker is forced to enable an executable stack for
executables that the code is linked into.
2014-03-11 14:24:16 +02:00
Martin Storsjö
c011890764 Push clobbered neon registers on the stack
According to the calling convention, the registers q4-q7 should be
preserved by functions. The caller (generated by the compiler) could
be using those registers anywhere for any intermediate data.

Functions that use more than 12 of the qX registers must push
the clobbered registers on the stack in order to be able to restore them
afterwards.

In functions that don't use all 16 registers, but clobber some of
the callee saved registers q4-q7, one or more of them are remapped
to reduce the number of registers that have to be saved/restored.

This incurs a very small (around 0.5%) slowdown in the decoder and
encoder.
2014-03-10 22:07:36 +02:00
Martin Storsjö
811c647c0e Remap registers to avoid clobbering the neon registers q4-q7
According to the calling convention, the registers q4-q7 should be
preserved by functions. The caller (generated by the compiler) could
be using those registers anywhere for any intermediate data.

Functions that use 12 or less of the qX registers can avoid
violating the calling convention by simply using other registers instead
of the callee saved registers q4-q7.

This change only remaps the registers used within functions - therefore
this does not affect performance at all. E.g. in functions using
registers q0-q7, we now use q0-q3 and q8-q11 instead.
2014-03-10 22:07:25 +02:00
Ethan Hugg
3627875986 Merge pull request from mstorsjo/use-common-threadlib
Make the processing lib use mutexes from WelsThreadLib from the common library
2014-03-10 09:45:51 -07:00
ruil2
44a49b1fef Merge pull request from mstorsjo/android-threading
Don't try to set thread scope and scheduling policy on android
2014-03-10 17:26:00 +08:00
ruil2
2539d6e447 Merge pull request from mstorsjo/fix-typos
Fix two typos in variable and macro names
2014-03-10 15:25:20 +08:00
Martin Storsjö
cc7b81f3c3 Fix a typo in arm assembly, LORD -> LOAD 2014-03-09 19:19:38 +02:00
Martin Storsjö
8d6b368a1c Remove unnecessary stray __cdecl annotations in function signature comments in x86 assembly 2014-03-09 19:18:02 +02:00
Martin Storsjö
1c6a910c11 Don't try to set thread scope and scheduling policy on android
These APIs aren't implemented on android.
2014-03-08 20:37:42 +02:00