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

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

Diego Biurrun authored Feb 05, 2017 and Henrik Gramner committed Aug 06, 2018

 * 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

eruffaldi authored Jan 06, 2018 and Henrik Gramner committed Aug 06, 2018

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

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

Anton Mitrofanov authored Apr 16, 2018 and Henrik Gramner committed Aug 06, 2018

Place it immediately after "static".

Anton Mitrofanov authored Jul 02, 2018 and Henrik Gramner committed Aug 06, 2018

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

Alexandra Hajkova authored Jul 08, 2018 and Henrik Gramner committed Aug 06, 2018

Increases overall encoding speed on POWER9 by 8%.

Alexandra Hájková authored Jul 11, 2018 and Henrik Gramner committed Aug 06, 2018

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

Anton Mitrofanov authored Sep 23, 2018 and Henrik Gramner committed Sep 25, 2018

Bug reported by Nicolas Gaullier

Anton Mitrofanov authored Oct 10, 2018 and Henrik Gramner committed Dec 23, 2018

Bug report by Koby Shina.

Anton Mitrofanov authored Nov 08, 2018 and Henrik Gramner committed Dec 23, 2018

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

Anton Mitrofanov authored Nov 08, 2018 and Henrik Gramner committed Dec 23, 2018

Bug report by Dirk Fieldhouse.

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

Anton Mitrofanov authored Feb 04, 2019 and Henrik Gramner committed Mar 03, 2019

Bug report by Daniel Deptford.

Anton Mitrofanov authored Feb 04, 2019 and Henrik Gramner committed Mar 03, 2019

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

Anton Mitrofanov authored Nov 09, 2018 and Henrik Gramner committed Mar 03, 2019

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