This work is sponsored by, and copyright, Google.
This avoids loading and calculating coefficients that we know will
be zero, and avoids filling the temp buffer with zeros in places
where we know the second pass won't read.
This gives a pretty substantial speedup for the smaller subpartitions.
The code size increases from 21512 bytes to 31400 bytes.
The idct16/32_end macros are moved above the individual functions; the
instructions themselves are unchanged, but since new functions are added
at the same place where the code is moved from, the diff looks rather
messy.
Before:
vp9_inv_dct_dct_16x16_sub1_add_10_neon: 284.6
vp9_inv_dct_dct_16x16_sub2_add_10_neon: 1902.7
vp9_inv_dct_dct_16x16_sub4_add_10_neon: 1903.0
vp9_inv_dct_dct_16x16_sub8_add_10_neon: 2201.1
vp9_inv_dct_dct_16x16_sub12_add_10_neon: 2510.0
vp9_inv_dct_dct_16x16_sub16_add_10_neon: 2821.3
vp9_inv_dct_dct_32x32_sub1_add_10_neon: 1011.6
vp9_inv_dct_dct_32x32_sub2_add_10_neon: 9716.5
vp9_inv_dct_dct_32x32_sub4_add_10_neon: 9704.9
vp9_inv_dct_dct_32x32_sub8_add_10_neon: 10641.7
vp9_inv_dct_dct_32x32_sub12_add_10_neon: 11555.7
vp9_inv_dct_dct_32x32_sub16_add_10_neon: 12499.8
vp9_inv_dct_dct_32x32_sub20_add_10_neon: 13403.7
vp9_inv_dct_dct_32x32_sub24_add_10_neon: 14335.8
vp9_inv_dct_dct_32x32_sub28_add_10_neon: 15253.6
vp9_inv_dct_dct_32x32_sub32_add_10_neon: 16179.5
After:
vp9_inv_dct_dct_16x16_sub1_add_10_neon: 282.8
vp9_inv_dct_dct_16x16_sub2_add_10_neon: 1142.4
vp9_inv_dct_dct_16x16_sub4_add_10_neon: 1139.0
vp9_inv_dct_dct_16x16_sub8_add_10_neon: 1772.9
vp9_inv_dct_dct_16x16_sub12_add_10_neon: 2515.2
vp9_inv_dct_dct_16x16_sub16_add_10_neon: 2823.5
vp9_inv_dct_dct_32x32_sub1_add_10_neon: 1012.7
vp9_inv_dct_dct_32x32_sub2_add_10_neon: 6944.4
vp9_inv_dct_dct_32x32_sub4_add_10_neon: 6944.2
vp9_inv_dct_dct_32x32_sub8_add_10_neon: 7609.8
vp9_inv_dct_dct_32x32_sub12_add_10_neon: 9953.4
vp9_inv_dct_dct_32x32_sub16_add_10_neon: 10770.1
vp9_inv_dct_dct_32x32_sub20_add_10_neon: 13418.8
vp9_inv_dct_dct_32x32_sub24_add_10_neon: 14330.7
vp9_inv_dct_dct_32x32_sub28_add_10_neon: 15257.1
vp9_inv_dct_dct_32x32_sub32_add_10_neon: 16190.6
Signed-off-by: Martin Storsjö <martin@martin.st>
This work is sponsored by, and copyright, Google.
This reduces the code size of libavcodec/aarch64/vp9itxfm_16bpp_neon.o from
26288 to 21512 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_10_neon: 1887.4
vp9_inv_dct_dct_16x16_sub16_add_10_neon: 2801.5
vp9_inv_dct_dct_32x32_sub4_add_10_neon: 9691.4
vp9_inv_dct_dct_32x32_sub32_add_10_neon: 16154.9
After:
vp9_inv_dct_dct_16x16_sub4_add_10_neon: 1899.5
vp9_inv_dct_dct_16x16_sub16_add_10_neon: 2827.2
vp9_inv_dct_dct_32x32_sub4_add_10_neon: 9714.7
vp9_inv_dct_dct_32x32_sub32_add_10_neon: 16175.9
Signed-off-by: Martin Storsjö <martin@martin.st>
Align the second/third operands as they usually are.
Due to the wildly varying sizes of the written out operands
in aarch64 assembly, the column alignment is usually not as clear
as in arm assembly.
This is cherrypicked from libav commit
7995ebfad1.
Signed-off-by: Martin Storsjö <martin@martin.st>
In the half/quarter cases where we don't use the min_eob array, defer
loading the pointer until we know it will be needed.
This is cherrypicked from libav commit
3a0d5e206d.
Signed-off-by: Martin Storsjö <martin@martin.st>
This matches the order they are in the 16 bpp version.
There they are in this order, to make sure we access them in the
same order they are declared, easing loading only half of the
coefficients at a time.
This makes the 8 bpp version match the 16 bpp version better.
This is cherrypicked from libav commit
b8f66c0838.
Signed-off-by: Martin Storsjö <martin@martin.st>
All elements are used pairwise, except for the first one.
Previously, the 16th element was unused. Move the unused element
to the second slot, to make the later element pairs not split
across registers.
This simplifies loading only parts of the coefficients,
reducing the difference to the 16 bpp version.
This is cherrypicked from libav commit
09eb88a12e.
Signed-off-by: Martin Storsjö <martin@martin.st>
The idct32x32 function actually pushed d8-d15 onto the stack even
though it didn't clobber them; there are plenty of registers that
can be used to allow keeping all the idct coefficients in registers
without having to reload different subsets of them at different
stages in the transform.
After this, we still can skip pushing d12-d15.
Before:
vp9_inv_dct_dct_32x32_sub32_add_neon: 8128.3
After:
vp9_inv_dct_dct_32x32_sub32_add_neon: 8053.3
This is cherrypicked from libav commit
65aa002d54.
Signed-off-by: Martin Storsjö <martin@martin.st>
This is one cycle faster in total, and three instructions fewer.
Before:
vp9_loop_filter_mix2_v_44_16_neon: 123.2
After:
vp9_loop_filter_mix2_v_44_16_neon: 122.2
This is cherrypicked from libav commit
3bf9c48320.
Signed-off-by: Martin Storsjö <martin@martin.st>
This adds lots of extra .ifs, but speeds it up by a couple cycles,
by avoiding stalls.
This is cherrypicked from libav commit
b0806088d3.
Signed-off-by: Martin Storsjö <martin@martin.st>
Previously we first calculated hev, and then negated it.
Since we were able to schedule the negation in the middle
of another calculation, we don't see any gain in all cases.
Before: Cortex A7 A8 A9 A53 A53/AArch64
vp9_loop_filter_v_4_8_neon: 147.0 129.0 115.8 89.0 88.7
vp9_loop_filter_v_8_8_neon: 242.0 198.5 174.7 140.0 136.7
vp9_loop_filter_v_16_8_neon: 500.0 419.5 382.7 293.0 275.7
vp9_loop_filter_v_16_16_neon: 971.2 825.5 731.5 579.0 453.0
After:
vp9_loop_filter_v_4_8_neon: 143.0 127.7 114.8 88.0 87.7
vp9_loop_filter_v_8_8_neon: 241.0 197.2 173.7 140.0 136.7
vp9_loop_filter_v_16_8_neon: 497.0 419.5 379.7 293.0 275.7
vp9_loop_filter_v_16_16_neon: 965.2 818.7 731.4 579.0 452.0
This is cherrypicked from libav commit
e1f9de86f4.
Signed-off-by: Martin Storsjö <martin@martin.st>
This work is sponsored by, and copyright, Google.
Before: Cortex A53
vp9_inv_dct_dct_16x16_sub1_add_neon: 235.3
vp9_inv_dct_dct_32x32_sub1_add_neon: 555.1
After:
vp9_inv_dct_dct_16x16_sub1_add_neon: 180.2
vp9_inv_dct_dct_32x32_sub1_add_neon: 475.3
This is cherrypicked from libav commit
3fcf788fbb.
Signed-off-by: Martin Storsjö <martin@martin.st>
No measured speedup on a Cortex A53, but other cores might benefit.
This is cherrypicked from libav commit
388e0d2515.
Signed-off-by: Martin Storsjö <martin@martin.st>
Fold the field lengths into the macro.
This makes the macro invocations much more readable, when the
lines are shorter.
This also makes it easier to use only half the registers within
the macro.
This is cherrypicked from libav commit
5e0c2158fb.
Signed-off-by: Martin Storsjö <martin@martin.st>
The ld1r is a leftover from the arm version, where this trick is
beneficial on some cores.
Use a single-lane load where we don't need the semantics of ld1r.
This is cherrypicked from libav commit
ed8d293306.
Signed-off-by: Martin Storsjö <martin@martin.st>
This work is sponsored by, and copyright, Google.
This avoids loading and calculating coefficients that we know will
be zero, and avoids filling the temp buffer with zeros in places
where we know the second pass won't read.
This gives a pretty substantial speedup for the smaller subpartitions.
The code size increases from 14740 bytes to 24292 bytes.
The idct16/32_end macros are moved above the individual functions; the
instructions themselves are unchanged, but since new functions are added
at the same place where the code is moved from, the diff looks rather
messy.
Before:
vp9_inv_dct_dct_16x16_sub1_add_neon: 236.7
vp9_inv_dct_dct_16x16_sub2_add_neon: 1051.0
vp9_inv_dct_dct_16x16_sub4_add_neon: 1051.0
vp9_inv_dct_dct_16x16_sub8_add_neon: 1051.0
vp9_inv_dct_dct_16x16_sub12_add_neon: 1387.4
vp9_inv_dct_dct_16x16_sub16_add_neon: 1387.6
vp9_inv_dct_dct_32x32_sub1_add_neon: 554.1
vp9_inv_dct_dct_32x32_sub2_add_neon: 5198.5
vp9_inv_dct_dct_32x32_sub4_add_neon: 5198.6
vp9_inv_dct_dct_32x32_sub8_add_neon: 5196.3
vp9_inv_dct_dct_32x32_sub12_add_neon: 6183.4
vp9_inv_dct_dct_32x32_sub16_add_neon: 6174.3
vp9_inv_dct_dct_32x32_sub20_add_neon: 7151.4
vp9_inv_dct_dct_32x32_sub24_add_neon: 7145.3
vp9_inv_dct_dct_32x32_sub28_add_neon: 8119.3
vp9_inv_dct_dct_32x32_sub32_add_neon: 8118.7
After:
vp9_inv_dct_dct_16x16_sub1_add_neon: 236.7
vp9_inv_dct_dct_16x16_sub2_add_neon: 640.8
vp9_inv_dct_dct_16x16_sub4_add_neon: 639.0
vp9_inv_dct_dct_16x16_sub8_add_neon: 842.0
vp9_inv_dct_dct_16x16_sub12_add_neon: 1388.3
vp9_inv_dct_dct_16x16_sub16_add_neon: 1389.3
vp9_inv_dct_dct_32x32_sub1_add_neon: 554.1
vp9_inv_dct_dct_32x32_sub2_add_neon: 3685.5
vp9_inv_dct_dct_32x32_sub4_add_neon: 3685.1
vp9_inv_dct_dct_32x32_sub8_add_neon: 3684.4
vp9_inv_dct_dct_32x32_sub12_add_neon: 5312.2
vp9_inv_dct_dct_32x32_sub16_add_neon: 5315.4
vp9_inv_dct_dct_32x32_sub20_add_neon: 7154.9
vp9_inv_dct_dct_32x32_sub24_add_neon: 7154.5
vp9_inv_dct_dct_32x32_sub28_add_neon: 8126.6
vp9_inv_dct_dct_32x32_sub32_add_neon: 8127.2
This is cherrypicked from libav commit
a63da4511d.
Signed-off-by: Martin Storsjö <martin@martin.st>
This allows reusing the macro for a separate implementation of the
pass2 function.
This is cherrypicked from libav commit
79d332ebbd.
Signed-off-by: Martin Storsjö <martin@martin.st>
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
This is cherrypicked from libav commit
115476018d.
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.
This is cherrypicked from libav commit
58d87e0f49.
Signed-off-by: Martin Storsjö <martin@martin.st>
This work is sponsored by, and copyright, Google.
This is similar to the arm version, but due to the larger registers
on aarch64, we can do 8 pixels at a time for all filter sizes.
Examples of runtimes vs the 32 bit version, on a Cortex A53:
ARM AArch64
vp9_loop_filter_h_4_8_10bpp_neon: 213.2 172.6
vp9_loop_filter_h_8_8_10bpp_neon: 281.2 244.2
vp9_loop_filter_h_16_8_10bpp_neon: 657.0 444.5
vp9_loop_filter_h_16_16_10bpp_neon: 1280.4 877.7
vp9_loop_filter_mix2_h_44_16_10bpp_neon: 397.7 358.0
vp9_loop_filter_mix2_h_48_16_10bpp_neon: 465.7 429.0
vp9_loop_filter_mix2_h_84_16_10bpp_neon: 465.7 428.0
vp9_loop_filter_mix2_h_88_16_10bpp_neon: 533.7 499.0
vp9_loop_filter_mix2_v_44_16_10bpp_neon: 271.5 244.0
vp9_loop_filter_mix2_v_48_16_10bpp_neon: 330.0 305.0
vp9_loop_filter_mix2_v_84_16_10bpp_neon: 329.0 306.0
vp9_loop_filter_mix2_v_88_16_10bpp_neon: 386.0 365.0
vp9_loop_filter_v_4_8_10bpp_neon: 150.0 115.2
vp9_loop_filter_v_8_8_10bpp_neon: 209.0 175.5
vp9_loop_filter_v_16_8_10bpp_neon: 492.7 345.2
vp9_loop_filter_v_16_16_10bpp_neon: 951.0 682.7
This is significantly faster than the ARM version in almost
all cases except for the mix2 functions.
Based on START_TIMER/STOP_TIMER wrapping around a few individual
functions, the speedup vs C code is around 2-3x.
Signed-off-by: Martin Storsjö <martin@martin.st>
This work is sponsored by, and copyright, Google.
Compared to the arm version, on aarch64 we can keep the full 8x8
transform in registers, and for 16x16 and 32x32, we can process
it in slices of 4 pixels instead of 2.
Examples of runtimes vs the 32 bit version, on a Cortex A53:
ARM AArch64
vp9_inv_adst_adst_4x4_sub4_add_10_neon: 111.0 109.7
vp9_inv_adst_adst_8x8_sub8_add_10_neon: 914.0 733.5
vp9_inv_adst_adst_16x16_sub16_add_10_neon: 5184.0 3745.7
vp9_inv_dct_dct_4x4_sub1_add_10_neon: 65.0 65.7
vp9_inv_dct_dct_4x4_sub4_add_10_neon: 100.0 96.7
vp9_inv_dct_dct_8x8_sub1_add_10_neon: 111.0 119.7
vp9_inv_dct_dct_8x8_sub8_add_10_neon: 618.0 494.7
vp9_inv_dct_dct_16x16_sub1_add_10_neon: 295.1 284.6
vp9_inv_dct_dct_16x16_sub2_add_10_neon: 2303.2 1883.9
vp9_inv_dct_dct_16x16_sub8_add_10_neon: 2984.8 2189.3
vp9_inv_dct_dct_16x16_sub16_add_10_neon: 3890.0 2799.4
vp9_inv_dct_dct_32x32_sub1_add_10_neon: 1044.4 1012.7
vp9_inv_dct_dct_32x32_sub2_add_10_neon: 13333.7 9695.1
vp9_inv_dct_dct_32x32_sub16_add_10_neon: 18531.3 12459.8
vp9_inv_dct_dct_32x32_sub32_add_10_neon: 24470.7 16160.2
vp9_inv_wht_wht_4x4_sub4_add_10_neon: 83.0 79.7
The larger transforms are significantly faster than the corresponding
ARM versions.
The speedup vs C code is smaller than in 32 bit mode, probably
because the 64 bit intermediates in the C code can be expressed
more efficiently in aarch64.
Signed-off-by: Martin Storsjö <martin@martin.st>
This work is sponsored by, and copyright, Google.
This has mostly got the same differences to the 8 bit version as
in the arm version. For the horizontal filters, we do 16 pixels
in parallel as well. For the 8 pixel wide vertical filters, we can
accumulate 4 rows before storing, just as in the 8 bit version.
Examples of runtimes vs the 32 bit version, on a Cortex A53:
ARM AArch64
vp9_avg4_10bpp_neon: 35.7 30.7
vp9_avg8_10bpp_neon: 93.5 84.7
vp9_avg16_10bpp_neon: 324.4 296.6
vp9_avg32_10bpp_neon: 1236.5 1148.2
vp9_avg64_10bpp_neon: 4639.6 4571.1
vp9_avg_8tap_smooth_4h_10bpp_neon: 130.0 128.0
vp9_avg_8tap_smooth_4hv_10bpp_neon: 440.0 440.5
vp9_avg_8tap_smooth_4v_10bpp_neon: 114.0 105.5
vp9_avg_8tap_smooth_8h_10bpp_neon: 327.0 314.0
vp9_avg_8tap_smooth_8hv_10bpp_neon: 918.7 865.4
vp9_avg_8tap_smooth_8v_10bpp_neon: 330.0 300.2
vp9_avg_8tap_smooth_16h_10bpp_neon: 1187.5 1155.5
vp9_avg_8tap_smooth_16hv_10bpp_neon: 2663.1 2591.0
vp9_avg_8tap_smooth_16v_10bpp_neon: 1107.4 1078.3
vp9_avg_8tap_smooth_64h_10bpp_neon: 17754.6 17454.7
vp9_avg_8tap_smooth_64hv_10bpp_neon: 33285.2 33001.5
vp9_avg_8tap_smooth_64v_10bpp_neon: 16066.9 16048.6
vp9_put4_10bpp_neon: 25.5 21.7
vp9_put8_10bpp_neon: 56.0 52.0
vp9_put16_10bpp_neon/armv8: 183.0 163.1
vp9_put32_10bpp_neon/armv8: 678.6 563.1
vp9_put64_10bpp_neon/armv8: 2679.9 2195.8
vp9_put_8tap_smooth_4h_10bpp_neon: 120.0 118.0
vp9_put_8tap_smooth_4hv_10bpp_neon: 435.2 435.0
vp9_put_8tap_smooth_4v_10bpp_neon: 107.0 98.2
vp9_put_8tap_smooth_8h_10bpp_neon: 303.0 290.0
vp9_put_8tap_smooth_8hv_10bpp_neon: 893.7 828.7
vp9_put_8tap_smooth_8v_10bpp_neon: 305.5 263.5
vp9_put_8tap_smooth_16h_10bpp_neon: 1089.1 1059.2
vp9_put_8tap_smooth_16hv_10bpp_neon: 2578.8 2452.4
vp9_put_8tap_smooth_16v_10bpp_neon: 1009.5 933.5
vp9_put_8tap_smooth_64h_10bpp_neon: 16223.4 15918.6
vp9_put_8tap_smooth_64hv_10bpp_neon: 32153.0 31016.2
vp9_put_8tap_smooth_64v_10bpp_neon: 14516.5 13748.1
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 around 4-9x.
Signed-off-by: Martin Storsjö <martin@martin.st>
This work is sponsored by, and copyright, Google.
This is more in line with how it will be extended for more bitdepths.
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.
This is cherrypicked from libav commits
cad42fadcd and
a0c443a398.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
The clobbering tests in checkasm are only invoked when testing
correctness, so this bug didn't show up when benchmarking the
dc-only version.
This is cherrypicked from libav commit
4d960a1185.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
The latter is 1 cycle faster on a cortex-53 and since the operands are
bytewise (or larger) bitmask (impossible to overflow to zero) both are
equivalent.
This is cherrypicked from libav commit
e7ae8f7a71.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Since aarch64 has enough free general purpose registers use them to
branch to the appropiate storage code. 1-2 cycles faster for the
functions using loop_filter 8/16, ... on a cortex-a53. Mixed results
(up to 2 cycles faster/slower) on a cortex-a57.
This is cherrypicked from libav commit
d7595de0b2.
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
Prep work for the next commit, which will add a new FFT algorithm
which makes the iMDCT over 3x faster than it is currently (standalone,
the FFT is with some framesizes over 10x faster).
The new FFT algorithm uses the already thouroughly SIMD'd power of two
FFT which already has SIMD for AArch64, so users of that platform will
still see an improvement.
The previous FFT+SIMD was barely 2.5x faster than the C versions on these
platforms.
Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
The former is not an official pseudo instruction although gas and llvm's
internal assembler support it. Fixes a build error with xcode 6.2
reported by Memphiz on github.
Signed-off-by: Martin Storsjö <martin@martin.st>
(cherry picked from commit 6a62795d40)
Cherry pick Suggested-by: Martin Storsjö
This should fix the build failure on macosx
Signed-off-by: Michael Niedermayer <michael@niedermayer.cc>
James Almer
Ronald S. Bultje
Clément Bœsch
Martin Storsjö
Matthieu Bouron
Janne Grunau
Rostislav Pehlivanov