]> granicus.if.org Git - libx264/log
libx264
5 years agoFix integer overflow detected by UBSan in --weightp analysis
Anton Mitrofanov [Mon, 22 Apr 2019 18:41:43 +0000 (21:41 +0300)]
Fix integer overflow detected by UBSan in --weightp analysis

Bug report by Xuezhi Yan.

5 years agocheckasm: Fix heap-buffer-overflow read detected by ASan
Anton Mitrofanov [Fri, 12 Apr 2019 12:40:01 +0000 (15:40 +0300)]
checkasm: Fix heap-buffer-overflow read detected by ASan

5 years agoFix heap-buffer-overflow read detected by ASan with interlaced encoding
Anton Mitrofanov [Fri, 12 Apr 2019 12:38:08 +0000 (15:38 +0300)]
Fix heap-buffer-overflow read detected by ASan with interlaced encoding

Bug report by Hongxu Chen.

5 years agoCI: Bump macos target to darwin18
Konstantin Pavlov [Tue, 16 Jul 2019 19:38:32 +0000 (22:38 +0300)]
CI: Bump macos target to darwin18

5 years agoCI: Use a newer aarch64 image
Konstantin Pavlov [Tue, 16 Jul 2019 19:24:46 +0000 (22:24 +0300)]
CI: Use a newer aarch64 image

It now includes pkg-config, so lavf can be detected.

5 years agoAdded gitlab CI
Konstantin Pavlov [Fri, 5 Apr 2019 12:08:29 +0000 (15:08 +0300)]
Added gitlab CI

Supported targets:
 - debian amd64
 - debian aarch64
 - windows 32 bit
 - windows 64 bit
 - macos 64bit

The tests are ran on all supported targets (via wine on windows).

The release jobs are only available on master/stable branches in
videolan/x264 repository, and must be ran manually when a developer
wishes to upload the artifacts.

5 years agoFix warning in autocomplete.c when compiled with lavf
Henrik Gramner [Thu, 14 Mar 2019 13:31:22 +0000 (14:31 +0100)]
Fix warning in autocomplete.c when compiled with lavf

5 years agoRemove compatibility workarounds
Anton Mitrofanov [Mon, 5 Jun 2017 23:30:41 +0000 (02:30 +0300)]
Remove compatibility workarounds

This will break decoding with older versions of FFmpeg/Libav.

5 years agoRemove h->rc dereferencing where possible
Anton Mitrofanov [Fri, 9 Nov 2018 15:37:17 +0000 (18:37 +0300)]
Remove h->rc dereferencing where possible

5 years agox86inc: Add support for GFNI instructions
Henrik Gramner [Sat, 16 Feb 2019 20:02:01 +0000 (21:02 +0100)]
x86inc: Add support for GFNI instructions

5 years agox86inc: Improve warnings for use of unsupported instructions
Henrik Gramner [Sat, 16 Feb 2019 16:57:21 +0000 (17:57 +0100)]
x86inc: Improve warnings for use of unsupported instructions

Warn when the following are used without the appropriate cpuflag:
 * YMM and ZMM registers
 * 'pextrw' with a memory operand
 * GPR instruction set extensions

5 years agox86inc: Support N_PEXT bit on Mach-O
Henrik Gramner [Thu, 31 Jan 2019 19:42:32 +0000 (20:42 +0100)]
x86inc: Support N_PEXT bit on Mach-O

Allows for marking symbols as having limited global scope, similar to
using 'hidden' symbol visibility on ELF.

5 years agox86inc: Make 'non-adjacent' default in the TAIL_CALL macro
Henrik Gramner [Thu, 31 Jan 2019 19:21:43 +0000 (20:21 +0100)]
x86inc: Make 'non-adjacent' default in the TAIL_CALL macro

5 years agox86inc: Add x86-32 PIC support macros
Henrik Gramner [Thu, 31 Jan 2019 19:17:56 +0000 (20:17 +0100)]
x86inc: Add x86-32 PIC support macros

5 years agox86inc: Turn 'movsxd' into 'movifnidn' on x86-32
Henrik Gramner [Thu, 31 Jan 2019 19:11:01 +0000 (20:11 +0100)]
x86inc: Turn 'movsxd' into 'movifnidn' on x86-32

5 years agoBump dates to 2019
Henrik Gramner [Thu, 31 Jan 2019 19:08:40 +0000 (20:08 +0100)]
Bump dates to 2019

5 years agocli: Bash autocomplete support
Henrik Gramner [Sun, 1 Jul 2018 18:34:48 +0000 (20:34 +0200)]
cli: Bash autocomplete support

Allows for automatic command line completion for both options and values.

Options such as --input-csp and --input-fmt will dynamically retrieve
supported values from libavformat when compiled with lavf support.

Execute 'source tools/bash-autocomplete.sh' in bash to enable.

5 years agoSignal Progressive and Constrained profiles
Yusuke Nakamura [Mon, 9 Apr 2018 02:01:28 +0000 (11:01 +0900)]
Signal Progressive and Constrained profiles

Progressive High, Constrained High, and Progressive High 10.

Even in Main profile, constraint_set4_flag is now set to 1 if progressive,
and constraint_set5_flag is set to 1 if no B-slices are present.

5 years agoppc: Use xxpermdi in sad_x3/x4 and use macros to avoid redundant code
Alexandra Hájková [Sat, 8 Sep 2018 07:15:53 +0000 (07:15 +0000)]
ppc: Use xxpermdi in sad_x3/x4 and use macros to avoid redundant code

5 years agoppc: Use the vec_xst_len for partial stores in mc
Luca Barbato [Thu, 6 Sep 2018 10:25:14 +0000 (12:25 +0200)]
ppc: Use the vec_xst_len for partial stores in mc

Around a ~1% speedup to the overall encoding for --slow.

5 years agoppc: Use vec_splats in mc
Luca Barbato [Thu, 6 Sep 2018 10:25:13 +0000 (12:25 +0200)]
ppc: Use vec_splats in mc

No overall speedup, just tidier code.

5 years agoppc: Use the vec_xst_len for partial stores
Luca Barbato [Thu, 23 Aug 2018 08:30:37 +0000 (08:30 +0000)]
ppc: Use the vec_xst_len for partial stores

Seems to give about a 1-2% overall speedup on --slow.

5 years agoppc: Use xxpermdi in VEC_STORE8
Luca Barbato [Sun, 19 Aug 2018 15:27:55 +0000 (17:27 +0200)]
ppc: Use xxpermdi in VEC_STORE8

Around a ~2% speedup to the overall encoding for --slow.

5 years agoppc: Use a single store to write the scores for sad_x4_8x8
Luca Barbato [Sun, 19 Aug 2018 15:27:54 +0000 (17:27 +0200)]
ppc: Use a single store to write the scores for sad_x4_8x8

Yet another use of xxpermdi, another 10% gain.

5 years agoppc: Use xxpermdi to halve the computation in sad_x4_8x8
Luca Barbato [Sun, 19 Aug 2018 15:27:53 +0000 (17:27 +0200)]
ppc: Use xxpermdi to halve the computation in sad_x4_8x8

About 20% faster.

5 years agoppc: Rework satd_4* likewise
Luca Barbato [Sun, 19 Aug 2018 07:28:42 +0000 (09:28 +0200)]
ppc: Rework satd_4* likewise

Now 4x4 is as slow as C and 4x8 is a 2% faster than before.

5 years agoppc: Factor out the sum of absolute
Luca Barbato [Sun, 19 Aug 2018 07:28:41 +0000 (09:28 +0200)]
ppc: Factor out the sum of absolute

And use it on the other satd > 8.

5-10% faster depending on the size.

5 years agoppc: Rework the adds in satd8x8
Luca Barbato [Sun, 19 Aug 2018 07:28:40 +0000 (09:28 +0200)]
ppc: Rework the adds in satd8x8

10% faster.

5 years agoppc: Add quant_4x4x4
Luca Barbato [Fri, 17 Aug 2018 20:28:45 +0000 (22:28 +0200)]
ppc: Add quant_4x4x4

4x faster than C.

5 years agoppc: Cleanup quant
Luca Barbato [Fri, 17 Aug 2018 20:28:44 +0000 (22:28 +0200)]
ppc: Cleanup quant

5 years agox86: Always use PIC in x86-64 asm
Henrik Gramner [Sun, 12 Aug 2018 15:00:13 +0000 (17:00 +0200)]
x86: Always use PIC in x86-64 asm

Most x86-64 operating systems nowadays doesn't even allow .text relocations
in object files any more, and there is no measurable overall performance
difference from using RIP-relative addressing in x264 asm.

Enforcing PIC reduces complexity and simplifies testing.

5 years agox86: Fix integer overflow in intra_sa8d_x3_8x8_sse2
Henrik Gramner [Sat, 23 Feb 2019 19:15:33 +0000 (20:15 +0100)]
x86: Fix integer overflow in intra_sa8d_x3_8x8_sse2

5 years agoCheck that mbtree settings are consistent between passes
Anton Mitrofanov [Fri, 9 Nov 2018 15:13:34 +0000 (18:13 +0300)]
Check that mbtree settings are consistent between passes

Also check that CQP mode is not used with 2-pass.

5 years agoMark frame_size_estimated as volatile
Anton Mitrofanov [Mon, 4 Feb 2019 19:04:56 +0000 (22:04 +0300)]
Mark frame_size_estimated as volatile

Ensures that access is atomic and that other threads sees the actual
value of the variable.

5 years agoFix data race detected by ThreadSanitizer
Anton Mitrofanov [Mon, 4 Feb 2019 18:46:12 +0000 (21:46 +0300)]
Fix data race detected by ThreadSanitizer

Bug report by Daniel Deptford.

5 years agoFix XAVC with sliced-threads
Anton Mitrofanov [Mon, 24 Dec 2018 16:37:45 +0000 (19:37 +0300)]
Fix XAVC with sliced-threads

5 years agoFix XAVC slice pattern
Anton Mitrofanov [Fri, 21 Dec 2018 15:54:56 +0000 (18:54 +0300)]
Fix XAVC slice pattern

5 years agoEliminate the use of strtok()
Henrik Gramner [Sun, 21 Oct 2018 12:28:59 +0000 (14:28 +0200)]
Eliminate the use of strtok()

Also fix the string parsing in param_apply_tune() to correctly compare
the entire string, not just the first N characters.

6 years agoconfigure: Fix log2f misdetection on some systems
Anton Mitrofanov [Thu, 8 Nov 2018 19:01:54 +0000 (22:01 +0300)]
configure: Fix log2f misdetection on some systems

Bug report by Dirk Fieldhouse.

6 years agoFix ultrafast preset speed regression
Anton Mitrofanov [Thu, 8 Nov 2018 18:53:17 +0000 (21:53 +0300)]
Fix ultrafast preset speed regression

--trellis 0 was missed for it during 8-bit and 10-bit unification.
Bug report by Aleksey Vasenev.

6 years agoFix --crop-rect top offset with --interlaced or --fake-interlaced
Anton Mitrofanov [Wed, 10 Oct 2018 16:41:08 +0000 (19:41 +0300)]
Fix --crop-rect top offset with --interlaced or --fake-interlaced

Bug report by Koby Shina.

6 years agoFix possible double transpose of custom CQM if --level is not set
Anton Mitrofanov [Sun, 23 Sep 2018 17:47:44 +0000 (20:47 +0300)]
Fix possible double transpose of custom CQM if --level is not set

Bug reported by Nicolas Gaullier

6 years agocli: Fix linking with --system-libx264 on x86
Henrik Gramner [Tue, 7 Aug 2018 20:42:22 +0000 (22:42 +0200)]
cli: Fix linking with --system-libx264 on x86

6 years agoFix CAVLC+RDO in 4:4:4
Anton Mitrofanov [Tue, 21 Aug 2018 12:11:21 +0000 (15:11 +0300)]
Fix CAVLC+RDO in 4:4:4

6 years agoppc: Optimize quant functions
Alexandra Hájková [Wed, 11 Jul 2018 19:28:20 +0000 (19:28 +0000)]
ppc: Optimize quant functions

1) using xxpermdi + merge instead of 2 merges improves quant_8x8
performance by 5%

2) use vec_splats instead of vec_splat

checkasm timings when compiled with gcc:
                  C:            AltiVec:
                                before: after:
quant_2x2_dc:      57            163      46
quant_4x4_dc:     141            162      57

dequant_4x4_cmp:  104            101      45
dequant_4x4_flat: 104            106      46
dequant_8x8_cmp:  412            208     147
dequant_8x8_flat: 414            212     149

6 years agoppc: Add support for Power9-only vec_absd
Alexandra Hajkova [Sun, 8 Jul 2018 18:04:43 +0000 (13:04 -0500)]
ppc: Add support for Power9-only vec_absd

Increases overall encoding speed on POWER9 by 8%.

6 years agoppc: Optimize sub8x8_dct_dc
Alexandra Hájková [Fri, 29 Jun 2018 16:50:20 +0000 (16:50 +0000)]
ppc: Optimize sub8x8_dct_dc

6 years agoppc: AltiVec add16x16_idct_dc
Alexandra Hájková [Thu, 21 Jun 2018 18:36:32 +0000 (18:36 +0000)]
ppc: AltiVec add16x16_idct_dc

6 years agoppc: Optimize add8x8_idct_dc
Alexandra Hájková [Sat, 23 Jun 2018 14:58:17 +0000 (14:58 +0000)]
ppc: Optimize add8x8_idct_dc

6 years agoppc: Add compatibility macros for vec_xxpermdi
Luca Barbato [Thu, 12 Jul 2018 08:41:22 +0000 (10:41 +0200)]
ppc: Add compatibility macros for vec_xxpermdi

6 years agoPrefer a monotonic clock source if available
Henrik Gramner [Sun, 24 Jun 2018 22:09:51 +0000 (00:09 +0200)]
Prefer a monotonic clock source if available

6 years agoAdd Sony XAVC, a flavour of AVC-Intra
Kieran Kunhya [Wed, 30 Aug 2017 15:05:41 +0000 (16:05 +0100)]
Add Sony XAVC, a flavour of AVC-Intra

6 years agoCosmetics: Fix indentation for multiline function prototypes
Anton Mitrofanov [Mon, 2 Jul 2018 17:20:03 +0000 (20:20 +0300)]
Cosmetics: Fix indentation for multiline function prototypes

It was broken in "Drop the x264 prefix" patch.

6 years agoCosmetics: Use consistent "inline" attribute position
Anton Mitrofanov [Mon, 16 Apr 2018 20:54:43 +0000 (23:54 +0300)]
Cosmetics: Use consistent "inline" attribute position

Place it immediately after "static".

6 years agox86: AVX-512 plane_copy and plane_copy_swap
Henrik Gramner [Thu, 25 Jan 2018 21:17:57 +0000 (22:17 +0100)]
x86: AVX-512 plane_copy and plane_copy_swap

Avoid the scalar C wrapper by utilizing opmasks to prevent overreading the
input buffer.

6 years ago4:0:0 (monochrome) encoding support
Emanuele Ruffaldi [Sat, 6 Jan 2018 01:34:39 +0000 (02:34 +0100)]
4:0:0 (monochrome) encoding support

Virtually zero increase in compression efficiency compared to 4:2:0 with empty
chroma planes. Performance is better though, especially with fast settings.

6 years agoMakefile improvements
Diego Biurrun [Sun, 5 Feb 2017 08:02:43 +0000 (09:02 +0100)]
Makefile improvements

 * Coalesce some install recipe lines

 * Remove empty addition of GPLed filters

 * Install libdir in recipes that directly require it

 * Coalesce etags/TAGS rules

 * Simplify fprofiled rule

6 years agox86inc: Improve SAVE/LOAD_MM_PERMUTATION macros
Henrik Gramner [Sun, 22 Apr 2018 20:49:15 +0000 (22:49 +0200)]
x86inc: Improve SAVE/LOAD_MM_PERMUTATION macros

Use register numbers instead of copying the full register names. This makes it
possible to change register widths in the middle of a function and keep the
mmreg permutations intact which can be useful for code that only needs larger
vectors for parts of the function in combination with macros etc.

Also change the LOAD_MM_PERMUTATION macro to use the same default name as the
SAVE macro. This simplifies swapping from ymm to xmm registers or vice versa:

    SAVE_MM_PERMUTATION
    INIT_XMM <cpuflags>
    LOAD_MM_PERMUTATION

6 years agox86inc: Optimize VEX instruction encoding
Henrik Gramner [Sat, 31 Mar 2018 11:49:56 +0000 (13:49 +0200)]
x86inc: Optimize VEX instruction encoding

Most VEX-encoded instructions require an additional byte to encode when src2
is a high register (e.g. x|ymm8..15). If the instruction is commutative we
can swap src1 and src2 when doing so reduces the instruction length, e.g.

    vpaddw xmm0, xmm0, xmm8 -> vpaddw xmm0, xmm8, xmm0

6 years agox86inc: Fix VEX -> EVEX instruction conversion
Henrik Gramner [Fri, 30 Mar 2018 23:16:06 +0000 (01:16 +0200)]
x86inc: Fix VEX -> EVEX instruction conversion

There's an edge case that wasn't properly handled.

6 years agoconfigure: Fix required version checks for lavf and swscale
Anton Mitrofanov [Tue, 31 Jul 2018 19:54:33 +0000 (22:54 +0300)]
configure: Fix required version checks for lavf and swscale

6 years agoFix float division by zero in weightp analysis
Anton Mitrofanov [Fri, 20 Jul 2018 05:37:43 +0000 (08:37 +0300)]
Fix float division by zero in weightp analysis

6 years agoFix undefined behavior of left shift for CAVLC encoding
Anton Mitrofanov [Wed, 18 Jul 2018 18:56:33 +0000 (21:56 +0300)]
Fix undefined behavior of left shift for CAVLC encoding

6 years agoFix integer overflow in slicetype_path_cost
Anton Mitrofanov [Mon, 2 Jul 2018 17:59:16 +0000 (20:59 +0300)]
Fix integer overflow in slicetype_path_cost

The path cost for high resolutions can exceed COST_MAX.

6 years agocli: Fix preset help listing
Henrik Gramner [Fri, 29 Jun 2018 11:14:01 +0000 (13:14 +0200)]
cli: Fix preset help listing

It was previously incorrect when --chroma-format or --bit-depth was
specified in configure.

6 years agoppc: Fix zigzag_interleave
Luca Barbato [Sat, 23 Jun 2018 11:14:28 +0000 (13:14 +0200)]
ppc: Fix zigzag_interleave

The permv array has 3 elements

6 years agoFix clang stack alignment issues
Henrik Gramner [Sat, 2 Jun 2018 18:35:10 +0000 (20:35 +0200)]
Fix clang stack alignment issues

Clang emits aligned AVX stores for things like zeroing stack-allocated
variables when using -mavx even with -fno-tree-vectorize set which can
result in crashes if this occurs before we've realigned the stack.

Previously we only ensured that the stack was realigned before calling
assembly functions that accesses stack-allocated buffers but this is
not sufficient. Fix the issue by changing the stack realignment to
instead occur immediately in all CLI, API and thread entry points.

6 years agoFix missing bs_flush in AUD writing
Anton Mitrofanov [Sun, 1 Apr 2018 17:49:29 +0000 (20:49 +0300)]
Fix missing bs_flush in AUD writing

6 years agoFix possible undefined behavior of right shift
Anton Mitrofanov [Sun, 1 Apr 2018 17:39:30 +0000 (20:39 +0300)]
Fix possible undefined behavior of right shift

32-bit shifts are only defined for values in the range 0-31.

6 years agoMake bs_align_10 imply bs_flush
Anton Mitrofanov [Sun, 1 Apr 2018 17:34:18 +0000 (20:34 +0300)]
Make bs_align_10 imply bs_flush

Now behaves the same as bs_align_0 and bs_align_1.

6 years agoFix theoretically incorrect cost_mv_fpel free
Anton Mitrofanov [Sun, 1 Apr 2018 14:52:47 +0000 (17:52 +0300)]
Fix theoretically incorrect cost_mv_fpel free

6 years agoconfigure: Fix ambiguous "$(("
Anton Mitrofanov [Sun, 1 Apr 2018 14:42:46 +0000 (17:42 +0300)]
configure: Fix ambiguous "$(("

6 years agoFix --qpmax default value in fullhelp
Anton Mitrofanov [Mon, 19 Feb 2018 16:53:38 +0000 (19:53 +0300)]
Fix --qpmax default value in fullhelp

6 years agox86: Correctly use v-prefix for instructions with opmasks
Henrik Gramner [Fri, 30 Mar 2018 23:31:57 +0000 (01:31 +0200)]
x86: Correctly use v-prefix for instructions with opmasks

This was always required, but accidentally happened to work correctly
in a few cases.

6 years agoconfigure: Only use gas-preprocessor with armasm for compiler=CL
Martin Storsjö [Fri, 30 Mar 2018 21:10:14 +0000 (00:10 +0300)]
configure: Only use gas-preprocessor with armasm for compiler=CL

This picks the right assembler automatically for arm and aarch64
llvm-mingw targets.

This doesn't get the right assembler for clang setups when clang
acts like MSVC and uses MSVC headers though (where it perhaps
should use armasm as before), but that's probably an even more
obscure setup.

6 years agoRemove ARRAY_SIZE macro which is identical to ARRAY_ELEMS
Anton Mitrofanov [Wed, 17 Jan 2018 19:03:06 +0000 (22:03 +0300)]
Remove ARRAY_SIZE macro which is identical to ARRAY_ELEMS

6 years agox86inc: Correctly set mmreg variables
Henrik Gramner [Sat, 6 Jan 2018 16:47:42 +0000 (17:47 +0100)]
x86inc: Correctly set mmreg variables

6 years ago.gitignore: Ignore TAGS file
Diego Biurrun [Sun, 5 Feb 2017 08:02:49 +0000 (09:02 +0100)]
.gitignore: Ignore TAGS file

6 years agoMinor configure improvements
Diego Biurrun [Sun, 5 Feb 2017 08:02:51 +0000 (09:02 +0100)]
Minor configure improvements

 * Drop empty addition of GPLed filters

 * Replace backticks with $()

6 years agoBump dates to 2018
Henrik Gramner [Mon, 1 Jan 2018 14:05:48 +0000 (15:05 +0100)]
Bump dates to 2018

6 years agoMerge zero buffers
Henrik Gramner [Tue, 16 Jan 2018 16:43:24 +0000 (17:43 +0100)]
Merge zero buffers

Improves cache efficiency.

6 years agordo: Use ALIGNED_ARRAY for stack arrays
Anton Mitrofanov [Wed, 17 Jan 2018 15:19:44 +0000 (18:19 +0300)]
rdo: Use ALIGNED_ARRAY for stack arrays

6 years agoCorrectly align buffers for AVX and AVX-512
Henrik Gramner [Mon, 15 Jan 2018 20:42:59 +0000 (21:42 +0100)]
Correctly align buffers for AVX and AVX-512

Fixes segfaults on Windows where the stack is only 16-byte aligned.

7 years agoCosmetics
Anton Mitrofanov [Sun, 24 Dec 2017 19:59:09 +0000 (22:59 +0300)]
Cosmetics

7 years agoppc: Add load_deinterleave_chroma_fenc_altivec
Alexandra Hájková [Sun, 21 May 2017 17:40:45 +0000 (17:40 +0000)]
ppc: Add load_deinterleave_chroma_fenc_altivec

5x speed up vs C code.

7 years agoUpdate to the latest upstream version of gas-preprocessor
Martin Storsjö [Thu, 26 Oct 2017 10:09:46 +0000 (13:09 +0300)]
Update to the latest upstream version of gas-preprocessor

This version supports converting aarch64 assembly for MS armasm64.exe.

7 years agoinput: Add a workaround for swscale overread bugs
Henrik Gramner [Sun, 22 Oct 2017 07:59:28 +0000 (09:59 +0200)]
input: Add a workaround for swscale overread bugs

swscale can read past the end of the input buffer, which may result in
crashes if such a read crosses a page boundary into an invalid page.

Work around this by adding some padding space at the end of the buffer when
using memory-mapped input frames. This may sometimes require copying the
last frame into a new buffer on Windows since the Microsoft memory-mapping
implementation has very limited capabilities compared to POSIX systems.

7 years agofilters/resize: Upgrade to a newer libavutil API
Henrik Gramner [Sun, 22 Oct 2017 08:50:46 +0000 (10:50 +0200)]
filters/resize: Upgrade to a newer libavutil API

Use the AVComponentDescriptor depth field instead of depth_minus1.

7 years agoaarch64: Use ldurb/sturb for loads/stores with negative offsets
Martin Storsjö [Wed, 18 Oct 2017 07:40:02 +0000 (10:40 +0300)]
aarch64: Use ldurb/sturb for loads/stores with negative offsets

The assembler (both gas and clang/llvm) automatically fixes this,
armasm64 doesn't. We can fix it in gas-preprocessor, but we should
also be using the right instruction form.

7 years agoconfigure: Add support for building with MSVC/armasm for ARM64
Martin Storsjö [Mon, 16 Oct 2017 19:50:27 +0000 (22:50 +0300)]
configure: Add support for building with MSVC/armasm for ARM64

7 years agoarm: Check for __ELF__ instead of !__APPLE__, for using .arch/.fpu
Martin Storsjö [Mon, 16 Oct 2017 19:50:26 +0000 (22:50 +0300)]
arm: Check for __ELF__ instead of !__APPLE__, for using .arch/.fpu

For windows, when building with armasm, we already filtered these out
with gas-preprocessor.

By filtering them out already in the source, we can also build directly
with clang for windows (which also require wrapping the assembler in
gas-preprocessor for converting instructions to thumb form, but
gas-preprocessor doesn't and shouldn't filter out them in the clang
configuration).

7 years agoaarch64: Don't .set a symbol named st2
Martin Storsjö [Mon, 16 Oct 2017 19:50:25 +0000 (22:50 +0300)]
aarch64: Don't .set a symbol named st2

This confuses gas-preprocessor, which tries to replace actual
st2 instructions by the integer 1 or 2.

7 years agoShrink the i4x4_mode cost_table array
Henrik Gramner [Sat, 14 Oct 2017 12:11:26 +0000 (14:11 +0200)]
Shrink the i4x4_mode cost_table array

Only 17 elements are actually used. It was originally padded to 64 bytes to
avoid cache line splits in the x86 assembly, but those haven't really been
an issue on x86 CPU:s made in the past decade or so.

Benchmarking shows no performance impact from dropping the padding, so
might as well remove it and save some cache.

7 years agox86: Remove some legacy CPU detection hacks
Henrik Gramner [Wed, 11 Oct 2017 16:02:26 +0000 (18:02 +0200)]
x86: Remove some legacy CPU detection hacks

Some ancient Pentium-M and Core 1 CPU:s had slow SSE units, and using MMX
was preferable. Nowadays many assembly functions in x264 completely lack MMX
implementations and falling back to C code will likely make things worse.

Some misconfigured virtualized systems could sometimes also trigger this code
path and cause assertions.

7 years agolavf: Upgrade to the new core decoding API
Henrik Gramner [Wed, 11 Oct 2017 15:58:36 +0000 (17:58 +0200)]
lavf: Upgrade to the new core decoding API

7 years agolavf: Upgrade to some newer API:s
Vittorio Giovara [Mon, 9 Oct 2017 16:04:22 +0000 (12:04 -0400)]
lavf: Upgrade to some newer API:s

 * Use the codec parameters API instead of the AVStream codec field.
 * Use av_packet_unref() instead of av_free_packet().
 * Use the AVFrame pts field instead of pkt_pts.

7 years agox86: AVX-512 load_deinterleave_chroma_fdec
Henrik Gramner [Sun, 8 Oct 2017 19:41:16 +0000 (21:41 +0200)]
x86: AVX-512 load_deinterleave_chroma_fdec

7 years agox86: AVX-512 load_deinterleave_chroma_fenc
Henrik Gramner [Sun, 8 Oct 2017 19:23:12 +0000 (21:23 +0200)]
x86: AVX-512 load_deinterleave_chroma_fenc

7 years agox86: AVX-512 mbtree_fix8_pack and mbtree_fix8_unpack
Henrik Gramner [Sat, 7 Oct 2017 10:06:51 +0000 (12:06 +0200)]
x86: AVX-512 mbtree_fix8_pack and mbtree_fix8_unpack

Takes advantage of opmasks to avoid having to use scalar code for the tail.

Also make some slight improvements to the checkasm test.

7 years agox86: Faster mbtree_fix8_unpack
Henrik Gramner [Sat, 7 Oct 2017 09:34:16 +0000 (11:34 +0200)]
x86: Faster mbtree_fix8_unpack

Use a different multiplier in order to eliminate some shifts.

About 25% faster than before.