This work is sponsored by, and copyright, Google.
This reduces the code size of libavcodec/aarch64/vp9itxfm_neon.o from
19496 to 14740 bytes.
This gives a small slowdown of a couple of tens of cycles, but makes
it more feasible to add more optimized versions of these transforms.
Before:
vp9_inv_dct_dct_16x16_sub4_add_neon: 1036.7
vp9_inv_dct_dct_16x16_sub16_add_neon: 1372.2
vp9_inv_dct_dct_32x32_sub4_add_neon: 5180.0
vp9_inv_dct_dct_32x32_sub32_add_neon: 8095.7
After:
vp9_inv_dct_dct_16x16_sub4_add_neon: 1051.0
vp9_inv_dct_dct_16x16_sub16_add_neon: 1390.1
vp9_inv_dct_dct_32x32_sub4_add_neon: 5199.9
vp9_inv_dct_dct_32x32_sub32_add_neon: 8125.8
Signed-off-by: Martin Storsjö <martin@martin.st>
This work is sponsored by, and copyright, Google.
This reduces the code size of libavcodec/arm/vp9itxfm_neon.o from
15324 to 12388 bytes.
This gives a small slowdown of a couple tens of cycles, up to around
150 cycles for the full case of the largest transform, but makes
it more feasible to add more optimized versions of these transforms.
Before: Cortex A7 A8 A9 A53
vp9_inv_dct_dct_16x16_sub4_add_neon: 2063.4 1516.0 1719.5 1245.1
vp9_inv_dct_dct_16x16_sub16_add_neon: 3279.3 2454.5 2525.2 1982.3
vp9_inv_dct_dct_32x32_sub4_add_neon: 10750.0 7955.4 8525.6 6754.2
vp9_inv_dct_dct_32x32_sub32_add_neon: 18574.0 17108.4 14216.7 12010.2
After:
vp9_inv_dct_dct_16x16_sub4_add_neon: 2060.8 1608.5 1735.7 1262.0
vp9_inv_dct_dct_16x16_sub16_add_neon: 3211.2 2443.5 2546.1 1999.5
vp9_inv_dct_dct_32x32_sub4_add_neon: 10682.0 8043.8 8581.3 6810.1
vp9_inv_dct_dct_32x32_sub32_add_neon: 18522.4 17277.4 14286.7 12087.9
Signed-off-by: Martin Storsjö <martin@martin.st>
This avoids having to count the number of frames sent to the codec
and the number of output packets received; instead just wait until
the encoder returns a buffer with the EOS flag set.
Signed-off-by: Martin Storsjö <martin@martin.st>
This avoids concatenation, which can't be used if the whole macro
is wrapped within another macro.
This is also arguably more readable.
Signed-off-by: Martin Storsjö <martin@martin.st>
This swaps which field is set when the Window Acknowledgement Size
and Set Peer BW packets are received, renames the fields in
order to clarify their role further and adds verbose comments
explaining their respective roles and how well the code currently
does what it is supposed to.
The Set Peer BW packet tells the receiver of the packet (which
can be either client or server) that it should not send more data
if it already has sent more data than the specified number of bytes,
without receiving acknowledgement for them. Actually checking this
limit is currently not implemented.
In order to be able to check that properly, one can send the
Window Acknowledgement Size packet, which tells the receiver of the
packet that it needs to send Acknowledgement packets
(RTMP_PT_BYTES_READ) at least after receiving a given number of bytes
since the last Acknowledgement.
Therefore, when we receive a Window Acknowledgement Size packet,
this sets the maximum number of bytes we can receive without sending
an Acknowledgement; therefore when handling this packet we should set
the receive_report_size field (previously client_report_size).
Signed-off-by: Martin Storsjö <martin@martin.st>
Also rename comments and log messages accordingly,
and add clarifying comments for some hardcoded values.
The previous names were taken from older, reverse engineered
references.
These names match the official public rtmp specification, and
matches the names used by wirecast in annotating captured
streams. These names also avoid hardcoding the roles of server
and client, since the handling of them is irrelevant of whether
we act as server or client.
The RTMP_PT_PING type maps to RTMP_PT_USER_CONTROL.
The SERVER_BW and CLIENT_BW types are a bit more intertwined;
RTMP_PT_SERVER_BW maps to RTMP_PT_WINDOW_ACK_SIZE and
RTMP_PT_CLIENT_BW maps to RTMP_PT_SET_PEER_BW.
Signed-off-by: Martin Storsjö <martin@martin.st>
If we only have a target compiler but no host compiler, the $type
variable will be empty once.
(Currently we fail to do a cross build if no host compiler is available
due to using the host compiler for processing option lists though.
But despite that, this comparison in configure needs quotes.)
Signed-off-by: Martin Storsjö <martin@martin.st>
When building DLLs with MSVC, CONFIG_STATIC is disabled (see
d66c52c2b3 for a more verbose explanation) since the built
object files can't be linked statically (which checkasm does).
This worked up until recently, only by luck.
Signed-off-by: Martin Storsjö <martin@martin.st>
Use a tab instead of two spaces, skip the fate prefix for the test name.
This makes IGNORE line fit in even better with the other make printouts.
Signed-off-by: Martin Storsjö <martin@martin.st>
Otherwise the .rep file would still contain a signal instead of a
zero, even if the process returned success.
Signed-off-by: Martin Storsjö <martin@martin.st>
A negative chunk size is illegal and would end up used as
length for memcpy, where it would lead to memory accesses
out of bounds.
Found-by: Paul Cher <paulcher@icloud.com>
CC: libav-stable@libav.org
Signed-off-by: Martin Storsjö <martin@martin.st>
This seems to have been added by mistake in 11de006b, by not
noticing the negation for the existing condition. This block does
not contain any code that accesses the codec field in AVStream.
This function is meant to serve as a complement to compute_pkt_fields2,
which is guarded by FF_API_COMPUTE_PKT_FIELDS2 && FF_API_LAVF_AVCTX.
Signed-off-by: Martin Storsjö <martin@martin.st>
This work is sponsored by, and copyright, Google.
Previously all subpartitions except the eob=1 (DC) case ran with
the same runtime:
vp9_inv_dct_dct_16x16_sub16_add_neon: 1373.2
vp9_inv_dct_dct_32x32_sub32_add_neon: 8089.0
By skipping individual 8x16 or 8x32 pixel slices in the first pass,
we reduce the runtime of these functions like this:
vp9_inv_dct_dct_16x16_sub1_add_neon: 235.3
vp9_inv_dct_dct_16x16_sub2_add_neon: 1036.7
vp9_inv_dct_dct_16x16_sub4_add_neon: 1036.7
vp9_inv_dct_dct_16x16_sub8_add_neon: 1036.7
vp9_inv_dct_dct_16x16_sub12_add_neon: 1372.1
vp9_inv_dct_dct_16x16_sub16_add_neon: 1372.1
vp9_inv_dct_dct_32x32_sub1_add_neon: 555.1
vp9_inv_dct_dct_32x32_sub2_add_neon: 5190.2
vp9_inv_dct_dct_32x32_sub4_add_neon: 5180.0
vp9_inv_dct_dct_32x32_sub8_add_neon: 5183.1
vp9_inv_dct_dct_32x32_sub12_add_neon: 6161.5
vp9_inv_dct_dct_32x32_sub16_add_neon: 6155.5
vp9_inv_dct_dct_32x32_sub20_add_neon: 7136.3
vp9_inv_dct_dct_32x32_sub24_add_neon: 7128.4
vp9_inv_dct_dct_32x32_sub28_add_neon: 8098.9
vp9_inv_dct_dct_32x32_sub32_add_neon: 8098.8
I.e. in general a very minor overhead for the full subpartition case due
to the additional cmps, but a significant speedup for the cases when we
only need to process a small part of the actual input data.
Signed-off-by: Martin Storsjö <martin@martin.st>
This work is sponsored by, and copyright, Google.
Previously all subpartitions except the eob=1 (DC) case ran with
the same runtime:
Cortex A7 A8 A9 A53
vp9_inv_dct_dct_16x16_sub16_add_neon: 3188.1 2435.4 2499.0 1969.0
vp9_inv_dct_dct_32x32_sub32_add_neon: 18531.7 16582.3 14207.6 12000.3
By skipping individual 4x16 or 4x32 pixel slices in the first pass,
we reduce the runtime of these functions like this:
vp9_inv_dct_dct_16x16_sub1_add_neon: 274.6 189.5 211.7 235.8
vp9_inv_dct_dct_16x16_sub2_add_neon: 2064.0 1534.8 1719.4 1248.7
vp9_inv_dct_dct_16x16_sub4_add_neon: 2135.0 1477.2 1736.3 1249.5
vp9_inv_dct_dct_16x16_sub8_add_neon: 2446.7 1828.7 1993.6 1494.7
vp9_inv_dct_dct_16x16_sub12_add_neon: 2832.4 2118.3 2266.5 1735.1
vp9_inv_dct_dct_16x16_sub16_add_neon: 3211.7 2475.3 2523.5 1983.1
vp9_inv_dct_dct_32x32_sub1_add_neon: 756.2 456.7 862.0 553.9
vp9_inv_dct_dct_32x32_sub2_add_neon: 10682.2 8190.4 8539.2 6762.5
vp9_inv_dct_dct_32x32_sub4_add_neon: 10813.5 8014.9 8518.3 6762.8
vp9_inv_dct_dct_32x32_sub8_add_neon: 11859.6 9313.0 9347.4 7514.5
vp9_inv_dct_dct_32x32_sub12_add_neon: 12946.6 10752.4 10192.2 8280.2
vp9_inv_dct_dct_32x32_sub16_add_neon: 14074.6 11946.5 11001.4 9008.6
vp9_inv_dct_dct_32x32_sub20_add_neon: 15269.9 13662.7 11816.1 9762.6
vp9_inv_dct_dct_32x32_sub24_add_neon: 16327.9 14940.1 12626.7 10516.0
vp9_inv_dct_dct_32x32_sub28_add_neon: 17462.7 15776.1 13446.2 11264.7
vp9_inv_dct_dct_32x32_sub32_add_neon: 18575.5 17157.0 14249.3 12015.1
I.e. in general a very minor overhead for the full subpartition case due
to the additional loads and cmps, but a significant speedup for the cases
when we only need to process a small part of the actual input data.
In common VP9 content in a few inspected clips, 70-90% of the non-dc-only
16x16 and 32x32 IDCTs only have nonzero coefficients in the upper left
8x8 or 16x16 subpartitions respectively.
Signed-off-by: Martin Storsjö <martin@martin.st>
This avoids reloading them if they haven't been clobbered, if the
first pass also was idct.
This is similar to what was done in the aarch64 version.
Signed-off-by: Martin Storsjö <martin@martin.st>
The rtmpdh code can use crypto libraries which may require
a process global init. (gcrypt is one of the libraries
where the rtmpdh test code can fail if global init hasn't been
done, depending on gcrypt version.)
Signed-off-by: Martin Storsjö <martin@martin.st>
Since the same parameter is used for both input and output,
the name inout is more fitting.
This matches the naming used below in the dmbutterfly macro.
Signed-off-by: Martin Storsjö <martin@martin.st>
This reverts commit 81d7f0bbca.
Instead of just benchmarking dc separately, test all relevant subparts
(in the next commit).
Signed-off-by: Martin Storsjö <martin@martin.st>
If makeopts_fate is set, these makeopts are used for running the
tests instead of the normal makeopts. If it isn't set, the normal
makeopts variable is used as before.
This is useful if remote testing on a lesser machine where a large
number of parallel jobs might be undesireable, while wanting to speed
up the build with many parallel processes.
Signed-off-by: Martin Storsjö <martin@martin.st>
The clobbering tests in checkasm are only invoked when testing
correctness, so this bug didn't show up when benchmarking the
dc-only version.
Signed-off-by: Martin Storsjö <martin@martin.st>
The dc-only mode is already checked to work correctly above, but this
allows benchmarking this mode for performance tuning, and allows making
sure that it actually is correctly hooked up.
Signed-off-by: Martin Storsjö <martin@martin.st>
This work is sponsored by, and copyright, Google.
These are ported from the ARM version; thanks to the larger
amount of registers available, we can do the loop filters with
16 pixels at a time. The implementation is fully templated, with
a single macro which can generate versions for both 8 and
16 pixels wide, for both 4, 8 and 16 pixels loop filters
(and the 4/8 mixed versions as well).
For the 8 pixel wide versions, it is pretty close in speed (the
v_4_8 and v_8_8 filters are the best examples of this; the h_4_8
and h_8_8 filters seem to get some gain in the load/transpose/store
part). For the 16 pixels wide ones, we get a speedup of around
1.2-1.4x compared to the 32 bit version.
Examples of runtimes vs the 32 bit version, on a Cortex A53:
ARM AArch64
vp9_loop_filter_h_4_8_neon: 144.0 127.2
vp9_loop_filter_h_8_8_neon: 207.0 182.5
vp9_loop_filter_h_16_8_neon: 415.0 328.7
vp9_loop_filter_h_16_16_neon: 672.0 558.6
vp9_loop_filter_mix2_h_44_16_neon: 302.0 203.5
vp9_loop_filter_mix2_h_48_16_neon: 365.0 305.2
vp9_loop_filter_mix2_h_84_16_neon: 365.0 305.2
vp9_loop_filter_mix2_h_88_16_neon: 376.0 305.2
vp9_loop_filter_mix2_v_44_16_neon: 193.2 128.2
vp9_loop_filter_mix2_v_48_16_neon: 246.7 218.4
vp9_loop_filter_mix2_v_84_16_neon: 248.0 218.5
vp9_loop_filter_mix2_v_88_16_neon: 302.0 218.2
vp9_loop_filter_v_4_8_neon: 89.0 88.7
vp9_loop_filter_v_8_8_neon: 141.0 137.7
vp9_loop_filter_v_16_8_neon: 295.0 272.7
vp9_loop_filter_v_16_16_neon: 546.0 453.7
The speedup vs C code in checkasm tests is around 2-7x, which is
pretty much the same as for the 32 bit version. Even if these functions
are faster than their 32 bit equivalent, the C version that we compare
to also became around 1.3-1.7x faster than the C version in 32 bit.
Based on START_TIMER/STOP_TIMER wrapping around a few individual
functions, the speedup vs C code is around 4-5x.
Examples of runtimes vs C on a Cortex A57 (for a slightly older version
of the patch):
A57 gcc-5.3 neon
loop_filter_h_4_8_neon: 256.6 93.4
loop_filter_h_8_8_neon: 307.3 139.1
loop_filter_h_16_8_neon: 340.1 254.1
loop_filter_h_16_16_neon: 827.0 407.9
loop_filter_mix2_h_44_16_neon: 524.5 155.4
loop_filter_mix2_h_48_16_neon: 644.5 173.3
loop_filter_mix2_h_84_16_neon: 630.5 222.0
loop_filter_mix2_h_88_16_neon: 697.3 222.0
loop_filter_mix2_v_44_16_neon: 598.5 100.6
loop_filter_mix2_v_48_16_neon: 651.5 127.0
loop_filter_mix2_v_84_16_neon: 591.5 167.1
loop_filter_mix2_v_88_16_neon: 855.1 166.7
loop_filter_v_4_8_neon: 271.7 65.3
loop_filter_v_8_8_neon: 312.5 106.9
loop_filter_v_16_8_neon: 473.3 206.5
loop_filter_v_16_16_neon: 976.1 327.8
The speed-up compared to the C functions is 2.5 to 6 and the cortex-a57
is again 30-50% faster than the cortex-a53.
Signed-off-by: Martin Storsjö <martin@martin.st>
This work is sponsored by, and copyright, Google.
These are ported from the ARM version; thanks to the larger
amount of registers available, we can do the 16x16 and 32x32
transforms in slices 8 pixels wide instead of 4. This gives
a speedup of around 1.4x compared to the 32 bit version.
The fact that aarch64 doesn't have the same d/q register
aliasing makes some of the macros quite a bit simpler as well.
Examples of runtimes vs the 32 bit version, on a Cortex A53:
ARM AArch64
vp9_inv_adst_adst_4x4_add_neon: 90.0 87.7
vp9_inv_adst_adst_8x8_add_neon: 400.0 354.7
vp9_inv_adst_adst_16x16_add_neon: 2526.5 1827.2
vp9_inv_dct_dct_4x4_add_neon: 74.0 72.7
vp9_inv_dct_dct_8x8_add_neon: 271.0 256.7
vp9_inv_dct_dct_16x16_add_neon: 1960.7 1372.7
vp9_inv_dct_dct_32x32_add_neon: 11988.9 8088.3
vp9_inv_wht_wht_4x4_add_neon: 63.0 57.7
The speedup vs C code (2-4x) is smaller than in the 32 bit case,
mostly because the C code ends up significantly faster (around
1.6x faster, with GCC 5.4) when built for aarch64.
Examples of runtimes vs C on a Cortex A57 (for a slightly older version
of the patch):
A57 gcc-5.3 neon
vp9_inv_adst_adst_4x4_add_neon: 152.2 60.0
vp9_inv_adst_adst_8x8_add_neon: 948.2 288.0
vp9_inv_adst_adst_16x16_add_neon: 4830.4 1380.5
vp9_inv_dct_dct_4x4_add_neon: 153.0 58.6
vp9_inv_dct_dct_8x8_add_neon: 789.2 180.2
vp9_inv_dct_dct_16x16_add_neon: 3639.6 917.1
vp9_inv_dct_dct_32x32_add_neon: 20462.1 4985.0
vp9_inv_wht_wht_4x4_add_neon: 91.0 49.8
The asm is around factor 3-4 faster than C on the cortex-a57 and the asm
is around 30-50% faster on the a57 compared to the a53.
Signed-off-by: Martin Storsjö <martin@martin.st>
This work is sponsored by, and copyright, Google.
The implementation tries to have smart handling of cases
where no pixels need the full filtering for the 8/16 width
filters, skipping both calculation and writeback of the
unmodified pixels in those cases. The actual effect of this
is hard to test with checkasm though, since it tests the
full filtering, and the benefit depends on how many filtered
blocks use the shortcut.
Examples of relative speedup compared to the C version, from checkasm:
Cortex A7 A8 A9 A53
vp9_loop_filter_h_4_8_neon: 2.72 2.68 1.78 3.15
vp9_loop_filter_h_8_8_neon: 2.36 2.38 1.70 2.91
vp9_loop_filter_h_16_8_neon: 1.80 1.89 1.45 2.01
vp9_loop_filter_h_16_16_neon: 2.81 2.78 2.18 3.16
vp9_loop_filter_mix2_h_44_16_neon: 2.65 2.67 1.93 3.05
vp9_loop_filter_mix2_h_48_16_neon: 2.46 2.38 1.81 2.85
vp9_loop_filter_mix2_h_84_16_neon: 2.50 2.41 1.73 2.85
vp9_loop_filter_mix2_h_88_16_neon: 2.77 2.66 1.96 3.23
vp9_loop_filter_mix2_v_44_16_neon: 4.28 4.46 3.22 5.70
vp9_loop_filter_mix2_v_48_16_neon: 3.92 4.00 3.03 5.19
vp9_loop_filter_mix2_v_84_16_neon: 3.97 4.31 2.98 5.33
vp9_loop_filter_mix2_v_88_16_neon: 3.91 4.19 3.06 5.18
vp9_loop_filter_v_4_8_neon: 4.53 4.47 3.31 6.05
vp9_loop_filter_v_8_8_neon: 3.58 3.99 2.92 5.17
vp9_loop_filter_v_16_8_neon: 3.40 3.50 2.81 4.68
vp9_loop_filter_v_16_16_neon: 4.66 4.41 3.74 6.02
The speedup vs C code is around 2-6x. The numbers are quite
inconclusive though, since the checkasm test runs multiple filterings
on top of each other, so later rounds might end up with different
codepaths (different decisions on which filter to apply, based
on input pixel differences). Disabling the early-exit in the asm
doesn't give a fair comparison either though, since the C code
only does the necessary calcuations for each row.
Based on START_TIMER/STOP_TIMER wrapping around a few individual
functions, the speedup vs C code is around 4-9x.
This is pretty similar in runtime to the corresponding routines
in libvpx. (This is comparing vpx_lpf_vertical_16_neon,
vpx_lpf_horizontal_edge_8_neon and vpx_lpf_horizontal_edge_16_neon
to vp9_loop_filter_h_16_8_neon, vp9_loop_filter_v_16_8_neon
and vp9_loop_filter_v_16_16_neon - note that the naming of horizonal
and vertical is flipped between the libraries.)
In order to have stable, comparable numbers, the early exits in both
asm versions were disabled, forcing the full filtering codepath.
Cortex A7 A8 A9 A53
vp9_loop_filter_h_16_8_neon: 597.2 472.0 482.4 415.0
libvpx vpx_lpf_vertical_16_neon: 626.0 464.5 470.7 445.0
vp9_loop_filter_v_16_8_neon: 500.2 422.5 429.7 295.0
libvpx vpx_lpf_horizontal_edge_8_neon: 586.5 414.5 415.6 383.2
vp9_loop_filter_v_16_16_neon: 905.0 784.7 791.5 546.0
libvpx vpx_lpf_horizontal_edge_16_neon: 1060.2 751.7 743.5 685.2
Our version is consistently faster on on A7 and A53, marginally slower on
A8, and sometimes faster, sometimes slower on A9 (marginally slower in all
three tests in this particular test run).
Signed-off-by: Martin Storsjö <martin@martin.st>
This work is sponsored by, and copyright, Google.
For the transforms up to 8x8, we can fit all the data (including
temporaries) in registers and just do a straightforward transform
of all the data. For 16x16, we do a transform of 4x16 pixels in
4 slices, using a temporary buffer. For 32x32, we transform 4x32
pixels at a time, in two steps of 4x16 pixels each.
Examples of relative speedup compared to the C version, from checkasm:
Cortex A7 A8 A9 A53
vp9_inv_adst_adst_4x4_add_neon: 3.39 5.83 4.17 4.01
vp9_inv_adst_adst_8x8_add_neon: 3.79 4.86 4.23 3.98
vp9_inv_adst_adst_16x16_add_neon: 3.33 4.36 4.11 4.16
vp9_inv_dct_dct_4x4_add_neon: 4.06 6.16 4.59 4.46
vp9_inv_dct_dct_8x8_add_neon: 4.61 6.01 4.98 4.86
vp9_inv_dct_dct_16x16_add_neon: 3.35 3.44 3.36 3.79
vp9_inv_dct_dct_32x32_add_neon: 3.89 3.50 3.79 4.42
vp9_inv_wht_wht_4x4_add_neon: 3.22 5.13 3.53 3.77
Thus, the speedup vs C code is around 3-6x.
This is mostly marginally faster than the corresponding routines
in libvpx on most cores, tested with their 32x32 idct (compared to
vpx_idct32x32_1024_add_neon). These numbers are slightly in libvpx's
favour since their version doesn't clear the input buffer like ours
do (although the effect of that on the total runtime probably is
negligible.)
Cortex A7 A8 A9 A53
vp9_inv_dct_dct_32x32_add_neon: 18436.8 16874.1 14235.1 11988.9
libvpx vpx_idct32x32_1024_add_neon 20789.0 13344.3 15049.9 13030.5
Only on the Cortex A8, the libvpx function is faster. On the other cores,
ours is slightly faster even though ours has got source block clearing
integrated.
Signed-off-by: Martin Storsjö <martin@martin.st>
This fixes crashes since 557c1675cf in linux PIC builds.
Previously, movrelx silently used r12 as helper register, which
doesn't work when r12 is the destination register.
Signed-off-by: Martin Storsjö <martin@martin.st>
We reset .Lpic_gp to zero at the start of each function, which means
that the logic within movrelx for clearing gp when necessary will
be missed.
This fixes using movrelx in different functions with a different
helper register.
Signed-off-by: Martin Storsjö <martin@martin.st>
This work is sponsored by, and copyright, Google.
The speedup for the large horizontal filters is surprisingly
big on A7 and A53, while there's a minor slowdown (almost within
measurement noise) on A8 and A9.
Cortex A7 A8 A9 A53
orig:
vp9_put_8tap_smooth_64h_neon: 20270.0 14447.3 19723.9 10910.9
new:
vp9_put_8tap_smooth_64h_neon: 20165.8 14466.5 19730.2 10668.8
Signed-off-by: Martin Storsjö <martin@martin.st>
This work is sponsored by, and copyright, Google.
These are ported from the ARM version; it is essentially a 1:1
port with no extra added features, but with some hand tuning
(especially for the plain copy/avg functions). The ARM version
isn't very register starved to begin with, so there's not much
to be gained from having more spare registers here - we only
avoid having to clobber callee-saved registers.
Examples of runtimes vs the 32 bit version, on a Cortex A53:
ARM AArch64
vp9_avg4_neon: 27.2 23.7
vp9_avg8_neon: 56.5 54.7
vp9_avg16_neon: 169.9 167.4
vp9_avg32_neon: 585.8 585.2
vp9_avg64_neon: 2460.3 2294.7
vp9_avg_8tap_smooth_4h_neon: 132.7 125.2
vp9_avg_8tap_smooth_4hv_neon: 478.8 442.0
vp9_avg_8tap_smooth_4v_neon: 126.0 93.7
vp9_avg_8tap_smooth_8h_neon: 241.7 234.2
vp9_avg_8tap_smooth_8hv_neon: 690.9 646.5
vp9_avg_8tap_smooth_8v_neon: 245.0 205.5
vp9_avg_8tap_smooth_64h_neon: 11273.2 11280.1
vp9_avg_8tap_smooth_64hv_neon: 22980.6 22184.1
vp9_avg_8tap_smooth_64v_neon: 11549.7 10781.1
vp9_put4_neon: 18.0 17.2
vp9_put8_neon: 40.2 37.7
vp9_put16_neon: 97.4 99.5
vp9_put32_neon/armv8: 346.0 307.4
vp9_put64_neon/armv8: 1319.0 1107.5
vp9_put_8tap_smooth_4h_neon: 126.7 118.2
vp9_put_8tap_smooth_4hv_neon: 465.7 434.0
vp9_put_8tap_smooth_4v_neon: 113.0 86.5
vp9_put_8tap_smooth_8h_neon: 229.7 221.6
vp9_put_8tap_smooth_8hv_neon: 658.9 621.3
vp9_put_8tap_smooth_8v_neon: 215.0 187.5
vp9_put_8tap_smooth_64h_neon: 10636.7 10627.8
vp9_put_8tap_smooth_64hv_neon: 21076.8 21026.9
vp9_put_8tap_smooth_64v_neon: 9635.0 9632.4
These are generally about as fast as the corresponding ARM
routines on the same CPU (at least on the A53), in most cases
marginally faster.
The speedup vs C code is pretty much the same as for the 32 bit
case; on the A53 it's around 6-13x for ther larger 8tap filters.
The exact speedup varies a little, since the C versions generally
don't end up exactly as slow/fast as on 32 bit.
Signed-off-by: Martin Storsjö <martin@martin.st>
With apple tools, the linker fails with errors like these, if the
offset is negative:
ld: in section __TEXT,__text reloc 8: symbol index out of range for architecture arm64
Signed-off-by: Martin Storsjö <martin@martin.st>
In recent lld-link versions, this command prints the version to
stdout, but also prints an error to stderr:
$ lld-link -flavor gnu --version
LLD 4.0.0 (trunk 285641)
lld-link: error: no input files
lld-link: error: target emulation unknown: -m or at least one .o file required
Signed-off-by: Martin Storsjö <martin@martin.st>
This fixes errors like this when building non-pic binaries with armv6
as baseline:
Error: invalid literal constant: pool needs to be closer
Signed-off-by: Martin Storsjö <martin@martin.st>
The include of config.h was added in 2012 in 1d9c2dc8, due to
the use of CONFIG_SNOW_ENCODER ifdefs within options_table.h.
When the snow codec was dropped later (in a0c5917f8 in 2013),
this include no longer served any purpose.
options_table.h is included in builds for the host as well, when
building documentation. config.h should not be included in code
that is built for the host, since it can contain workarounds
for the target compiler/environment, like adding a missing define
of restrict, defining getenv(x) to NULL for environments that lack
getenv.
The seemingly innocent include reordering in 2025d37871 broke
builds that have getenv(x) defined to NULL in config.h (Windows CE
and Windows Phone/RT), since libavcodec/options_table.h include
config.h, while libavformat/options_table.h end up bringing in
more system headers, and those system headers can contain a proper
definition of getenv, which clash with the getenv define in config.h.
This was avoided earlier as long as libavformat/options_table.h (or
avformat.h) was included before libavcodec/options_table.h.
This fixes builds for Windows Phone/RT and CE.
Signed-off-by: Martin Storsjö <martin@martin.st>
This work is sponsored by, and copyright, Google.
The filter coefficients are signed values, where the product of the
multiplication with one individual filter coefficient doesn't
overflow a 16 bit signed value (the largest filter coefficient is
127). But when the products are accumulated, the resulting sum can
overflow the 16 bit signed range. Instead of accumulating in 32 bit,
we accumulate the largest product (either index 3 or 4) last with a
saturated addition.
(The VP8 MC asm does something similar, but slightly simpler, by
accumulating each half of the filter separately. In the VP9 MC
filters, each half of the filter can also overflow though, so the
largest component has to be handled individually.)
Examples of relative speedup compared to the C version, from checkasm:
Cortex A7 A8 A9 A53
vp9_avg4_neon: 1.71 1.15 1.42 1.49
vp9_avg8_neon: 2.51 3.63 3.14 2.58
vp9_avg16_neon: 2.95 6.76 3.01 2.84
vp9_avg32_neon: 3.29 6.64 2.85 3.00
vp9_avg64_neon: 3.47 6.67 3.14 2.80
vp9_avg_8tap_smooth_4h_neon: 3.22 4.73 2.76 4.67
vp9_avg_8tap_smooth_4hv_neon: 3.67 4.76 3.28 4.71
vp9_avg_8tap_smooth_4v_neon: 5.52 7.60 4.60 6.31
vp9_avg_8tap_smooth_8h_neon: 6.22 9.04 5.12 9.32
vp9_avg_8tap_smooth_8hv_neon: 6.38 8.21 5.72 8.17
vp9_avg_8tap_smooth_8v_neon: 9.22 12.66 8.15 11.10
vp9_avg_8tap_smooth_64h_neon: 7.02 10.23 5.54 11.58
vp9_avg_8tap_smooth_64hv_neon: 6.76 9.46 5.93 9.40
vp9_avg_8tap_smooth_64v_neon: 10.76 14.13 9.46 13.37
vp9_put4_neon: 1.11 1.47 1.00 1.21
vp9_put8_neon: 1.23 2.17 1.94 1.48
vp9_put16_neon: 1.63 4.02 1.73 1.97
vp9_put32_neon: 1.56 4.92 2.00 1.96
vp9_put64_neon: 2.10 5.28 2.03 2.35
vp9_put_8tap_smooth_4h_neon: 3.11 4.35 2.63 4.35
vp9_put_8tap_smooth_4hv_neon: 3.67 4.69 3.25 4.71
vp9_put_8tap_smooth_4v_neon: 5.45 7.27 4.49 6.52
vp9_put_8tap_smooth_8h_neon: 5.97 8.18 4.81 8.56
vp9_put_8tap_smooth_8hv_neon: 6.39 7.90 5.64 8.15
vp9_put_8tap_smooth_8v_neon: 9.03 11.84 8.07 11.51
vp9_put_8tap_smooth_64h_neon: 6.78 9.48 4.88 10.89
vp9_put_8tap_smooth_64hv_neon: 6.99 8.87 5.94 9.56
vp9_put_8tap_smooth_64v_neon: 10.69 13.30 9.43 14.34
For the larger 8tap filters, the speedup vs C code is around 5-14x.
This is significantly faster than libvpx's implementation of the same
functions, at least when comparing the put_8tap_smooth_64 functions
(compared to vpx_convolve8_horiz_neon and vpx_convolve8_vert_neon from
libvpx).
Absolute runtimes from checkasm:
Cortex A7 A8 A9 A53
vp9_put_8tap_smooth_64h_neon: 20150.3 14489.4 19733.6 10863.7
libvpx vpx_convolve8_horiz_neon: 52623.3 19736.4 21907.7 25027.7
vp9_put_8tap_smooth_64v_neon: 14455.0 12303.9 13746.4 9628.9
libvpx vpx_convolve8_vert_neon: 42090.0 17706.2 17659.9 16941.2
Thus, on the A9, the horizontal filter is only marginally faster than
libvpx, while our version is significantly faster on the other cores,
and the vertical filter is significantly faster on all cores. The
difference is especially large on the A7.
The libvpx implementation does the accumulation in 32 bit, which
probably explains most of the differences.
Signed-off-by: Martin Storsjö <martin@martin.st>
This makes it match the pattern already used for VP8 MC functions.
This also makes the signature match ffmpeg's version of these
functions, easing porting of code in both directions.
Signed-off-by: Martin Storsjö <martin@martin.st>
When the macro is expanded with a semicolon following it and the
macro itself contains a semicolon, we ended up in double semicolons,
which is treated as a statement that disallows further declarations.
This avoids errors about mixed declarations and statements on gcc,
after ee05079766.
Signed-off-by: Martin Storsjö <martin@martin.st>
Instead use our own struct, which we already use when using
gcrypt and gnutls.
In OpenSSL 1.1, the DH struct has been made opaque.
Signed-off-by: Martin Storsjö <martin@martin.st>
x29 (FP) is a callee saved register and should be restored on
return. Instead of backing up x29 and restoring it here, back up
sp in a register that we are allowed to overwrite.
This fixes crashes in checkasm on aarch64 since f1b3e13138.
For some reason, gcc builds didn't crash, but clang builds do.
Signed-off-by: Martin Storsjö <martin@martin.st>
This, combined with clobbering the stack space prior to the call,
increases the chances of finding cases where 32 bit parameters
are erroneously treated as 64 bit.
Signed-off-by: Martin Storsjö <martin@martin.st>
Even if MAX_ARGS - 2 (for arm) or MAX_ARGS - 7 (for aarch64) parameters
are passed on the stack to checkasm_checked_call, we actually only
need to store MAX_ARGS - 4 (for arm) or MAX_ARGS - 8 (for aarch64)
parameters on the stack when calling the tested function.
Signed-off-by: Martin Storsjö <martin@martin.st>
According to the public RTMP specification, these 4 bytes should
be zero.
librtmp in server mode assumes that the RTMPE (FP9) handshake is
used if these bytes are nonzero.
Signed-off-by: Martin Storsjö <martin@martin.st>
When acting as server, the server can include a "clientid" property
in some status messages. But this should be a unique number
identifying the client session, not identifying the server itself.
In practice, omitting it works just as well as including this
incorrect field.
Signed-off-by: Martin Storsjö <martin@martin.st>
This makes sure that e.g. Adobe FME actually reacts to it. As long
as the value we've been sending is the default one (128), the bug
hasn't been noticed.
Signed-off-by: Martin Storsjö <martin@martin.st>
Some applications such as Adobe FME append lots of parameters
here, making it easily overflow the current limit.
Signed-off-by: Martin Storsjö <martin@martin.st>
This was introduced in bc2a32969e.
The whole block that the statement was added to is only
relevant when used as a demuxer, but the other statements
there have had other if statements guarding them. Make
sure to only run this whole block if being used as a
demuxer.
CC: libav-stable@libav.org
Signed-off-by: Martin Storsjö <martin@martin.st>
This reverts commit 014773b66b.
Since 230b1c070, the bytewise AV_W*() macros only expand their
argument once, i.e. doing exactly the same change as was done
in the AV_COPY*U macros, so this change is no longer necessary.
Signed-off-by: Martin Storsjö <martin@martin.st>
This reverts commit 25bacd0a0c.
Since 230b1c070, the bytewise AV_W*() macros only expand their
argument once, so revert to the more readable version of these.
Signed-off-by: Martin Storsjö <martin@martin.st>
AV_WN64 is meant for unaligned data, but the existing av_alias* unions
(without a definition for the av_alias attribute - we don't have one
for MSVC) indicate to the compiler that they would have sufficient
alignment for normal access, i.e. the compiler is free to assume
8 byte alignment.
On ARM, this makes sure that AV_WN64 (or two consecutive AV_WN32) is
done with two str instructions instead of one strd.
Signed-off-by: Martin Storsjö <martin@martin.st>
This avoids issues with expanding the argument multiple times,
and makes sure that it is of the right type for the following shifts.
Even if the caller of a macro could be expected not to pass parameters
that have side effects if expanded multiple times, these fallback
codepaths are rarely, if ever, tested, so it is expected that such
issues can arise.
Thefore, for safety, make sure the fallback codepaths only expand
the arguments once.
Signed-off-by: Martin Storsjö <martin@martin.st>
If AV_RN and AV_WN are macros with multiple individual reads and
writes, the previous version of the AV_COPYU macro would fail if
the reads and writes overlap.
This should not be any less efficient in any case, given a
sensibly optimizing compiler.
Signed-off-by: Martin Storsjö <martin@martin.st>
AV_WB32 can be implemented as a macro that expands its parameters
multiple times (in case AV_HAVE_FAST_UNALIGNED isn't set and the
compiler doesn't support GCC attributes); make sure not to read
multiple times from the source in this case.
Signed-off-by: Martin Storsjö <martin@martin.st>
This allows doing this redirection, if building with clang against
old enough MSVC headers that lack strtoll (2012 and older).
Signed-off-by: Martin Storsjö <martin@martin.st>
Clang normally disguises as GCC (defining __GNUC__), and thus get
all the normal GCC specific attributes.
Clang can also work as a drop-in replacement for MSVC, and in these
cases, it doesn't define __GNUC__, but defines _MSC_VER instead.
Even in these setups, it still supports the GCC style attributes,
thus use them, especially where there isn't any MSVC specific
version, or where the MSVC specific version doesn't work on clang
(for DECLARE_ASM_CONST).
Signed-off-by: Martin Storsjö <martin@martin.st>
When targeting COFF (windows), clang doesn't support this
directive (while binutils supports it for all targets).
Signed-off-by: Martin Storsjö <martin@martin.st>
These bits are set by exceptions in NEON instructions.
Also print the differing bits when FPSCR is clobbered,
and use bic instead of lsl, for clearing the topmost bits.
Signed-off-by: Martin Storsjö <martin@martin.st>
Each const block needs to be terminated by one endconst
invocation so either call endconst after each, or just
declare plain labels to the later strings.
This fixes errors such as this, on some binutils versions:
checkasm.S:38: Error: Macro `endconst' was already defined
Signed-off-by: Martin Storsjö <martin@martin.st>
This comment isn't true, the height can be different from the width
for these functions (which is why the height is passed as a parameter
to them).
Signed-off-by: Martin Storsjö <martin@martin.st>
When writing a fragmented file, we by default write an index pointing
to all the fragments at the end of the file. This causes constantly
increasing memory usage during the muxing. For live streams, the
index might not be useful at all.
A similar fragment index is written (but at the start of the file) if
the global_sidx flag is set. If ism_lookahead is set, we need to keep
data about the last ism_lookahead+1 fragments.
If no fragment index is to be written, we don't need to store information
about all fragments, avoiding increasing the memory consumption
linearly with the muxing runtime.
This fixes out of memory situations with long live mp4 streams.
Signed-off-by: Martin Storsjö <martin@martin.st>
This avoids listing the same feature multiple times in the
test output. Previously the output contained something like this:
SSE2:
- hevc_mc.qpel [OK]
- hevc_mc.epel [OK]
- hevc_mc.unweighted_pred [OK]
- hevc_mc.qpel [OK]
- hevc_mc.epel [OK]
- hevc_mc.unweighted_pred [OK]
Signed-off-by: Martin Storsjö <martin@martin.st>
This avoids the risk of accidentally clobbering such variables outside
of the macro if the same variables are used there.
Signed-off-by: Martin Storsjö <martin@martin.st>
This fixes valgrind warnings about conditional jumps based on
uninitialized data (even though the uninitialized data only ever
was compared with a direct copy of the same uninitialized data).
Signed-off-by: Martin Storsjö <martin@martin.st>
While it is less featureful (and slower) than the built-in H264
decoder, one could potentially want to use it to take advantage
of the cisco patent license offer.
Signed-off-by: Martin Storsjö <martin@martin.st>
The encode function is supposed to just return 0 on success.
This stems from a mixup with the return value of decode functions.
Signed-off-by: Martin Storsjö <martin@martin.st>
The functions may not clean up properly after using MMX
registers. For the normal testing calls, the checkasm_checked_call
functions will do the cleanup (and check that functions that
should clean up do it as well), but when benchmarking functions
that don't clean up, we don't currently properly clean up at all.
This causes issues if a benchmarked function is followed by testing
of a function that is supposed to not clobber the MMX/FPU state but
doesn't touch it at all.
Signed-off-by: Martin Storsjö <martin@martin.st>
Martin Storsjö
