clang-8:
cdef_filter_4x4_8bpc_c: 436.6
cdef_filter_4x4_8bpc_vsx: 101.1
cdef_filter_4x8_8bpc_c: 827.7
cdef_filter_4x8_8bpc_vsx: 183.5
cdef_filter_8x8_8bpc_c: 1510.2
cdef_filter_8x8_8bpc_vsx: 289.1

gcc-9:
cdef_filter_4x4_8bpc_c: 403.2
cdef_filter_4x4_8bpc_vsx: 105.6
cdef_filter_4x8_8bpc_c: 825.5
cdef_filter_4x8_8bpc_vsx: 192.2
cdef_filter_8x8_8bpc_c: 1586.3
cdef_filter_8x8_8bpc_vsx: 295.0

sdl2.pc adds -I${includedir}/SDL2 to the command line, so SDL2/ is
clearly expected.

Fixes #289

B Krishnan Iyer authored 5 years ago and Martin Storsjö committed 5 years ago

	                        A73		A53
	                Current	Earlier	Current	Earlier
blend_h_w2_8bpc_neon:	71.1	74.1	132.7	137.5
blend_h_w4_8bpc_neon:	60.2	65.8	137.5	147.1
blend_h_w8_8bpc_neon:	62.2	68.9	123.1	131.7
blend_h_w16_8bpc_neon:	82.1	86	180.7	190.3
blend_h_w32_8bpc_neon:	149.9	149.2	358.3	358
blend_h_w64_8bpc_neon:	265.3	263.1	630.2	629.8
blend_h_w128_8bpc_neon:	579.5	571	1404.4	1404.5
blend_v_w2_8bpc_neon:	118.7	118.7	193.2	195.3
blend_v_w4_8bpc_neon:	248.6	245.8	373.4	357.3
blend_v_w8_8bpc_neon:	202.7	202	356.4	357.2
blend_v_w16_8bpc_neon:	238.8	234.8	590.4	591.3
blend_v_w32_8bpc_neon:	346.7	344.4	993.7	994.7
blend_w4_8bpc_neon:	33.5	37.5	90.7	96.7
blend_w8_8bpc_neon:	49.7	53	123.3	123.3
blend_w16_8bpc_neon:	151.8	151	348.8	332.4
blend_w32_8bpc_neon:	372.9	370.9	908.3	908.4

B Krishnan Iyer authored 5 years ago and Martin Storsjö committed 5 years ago

	                	A73		A53
                	Current	Earlier	Current	Earlier
blend_h_w2_8bpc_neon:	74.1	74.1	137.5	137.5
blend_h_w4_8bpc_neon:	65.8	65.8	147.1	147.1
blend_h_w8_8bpc_neon:	68.9	68.7	131.7	131.7
blend_h_w16_8bpc_neon:	86	85.6	190.3	190.4
blend_h_w32_8bpc_neon:	149.2	149.8	358	358.3
blend_h_w64_8bpc_neon:	263.1	264.1	629.8	630.3
blend_h_w128_8bpc_neon:	571	575.4	1404.5	1404.2
blend_v_w2_8bpc_neon:	118.7	120.1	195.3	196.4
blend_v_w4_8bpc_neon:	245.8	247.2	357.3	358.4
blend_v_w8_8bpc_neon:	202	204.2	357.2	358.4
blend_v_w16_8bpc_neon:	234.8	238.5	591.3	591.8
blend_v_w32_8bpc_neon:	344.4	347.2	994.7	997.2
blend_w4_8bpc_neon:	37.5	38.3	96.7	98.7
blend_w8_8bpc_neon:	53	54.8	123.3	125.3
blend_w16_8bpc_neon:	151	150.8	332.4	334.5
blend_w32_8bpc_neon:	370.9	361.6	908.4	910.7

B Krishnan Iyer authored 5 years ago and Martin Storsjö committed 5 years ago

arm: mc: neon: Use vld with ! post-increment instead of a register in blend/blend_h/blend_v function

	                        A73		A53
	                Current	Earlier	Current	Earlier
blend_h_w2_8bpc_neon:	74.1	74.6	137.5	137
blend_h_w4_8bpc_neon:	65.8	66	147.1	146.6
blend_h_w8_8bpc_neon:	68.7	68.6	131.7	131.2
blend_h_w16_8bpc_neon:	85.6	85.9	190.4	192
blend_h_w32_8bpc_neon:	149.8	149.8	358.3	357.6
blend_h_w64_8bpc_neon:	264.1	262.8	630.3	629.5
blend_h_w128_8bpc_neon:	575.4	577	1404.2	1402
blend_v_w2_8bpc_neon:	120.1	121.3	196.4	195.5
blend_v_w4_8bpc_neon:	247.2	247.5	358.4	358.5
blend_v_w8_8bpc_neon:	204.2	205.2	358.4	358.5
blend_v_w16_8bpc_neon:	238.5	237.1	591.8	590.5
blend_v_w32_8bpc_neon:	347.2	345.8	997.2	994.1
blend_w4_8bpc_neon:	38.3	38.6	98.7	99.2
blend_w8_8bpc_neon:	54.8	55.1	125.3	125.8
blend_w16_8bpc_neon:	150.8	150.1	334.5	344
blend_w32_8bpc_neon:	361.6	360.4	910.7	910.9

Henrik Gramner authored 5 years ago and Henrik Gramner committed 5 years ago

Continuing trying to decode after a memory allocation failure could
cause null pointer dereferences in certain scenarios.

Emmanuel Gil Peyrot authored 5 years ago and Jean-Baptiste Kempf committed 5 years ago

This is using the Linux-only prctl(PR_SET_NAME, …) call, because glibc’s
pthread_setname_np() is doing exactly the same call so there is no
reason to use it instead, as it isn’t any more portable.

I don’t have any other OS to test this on, but if you want to add one
just add an #else defined(__YOUR_OS__) before the #else in thread.h.

		                        A73		A53

w_mask_420_w4_8bpc_c:	        	797.5		1072.7
w_mask_420_w4_8bpc_neon:		85.6		152.7
w_mask_420_w8_8bpc_c:		        2344.3		3118.7
w_mask_420_w8_8bpc_neon:		221.9		372.4
w_mask_420_w16_8bpc_c:		        7429.9		9702.1
w_mask_420_w16_8bpc_neon:		620.4		1024.1
w_mask_420_w32_8bpc_c:	        	27498.2		37205.7
w_mask_420_w32_8bpc_neon:		2394.1		3838
w_mask_420_w64_8bpc_c:  		66495.8		88721.3
w_mask_420_w64_8bpc_neon:      		6081.4		9630
w_mask_420_w128_8bpc_c:	        	163369.3	219494
w_mask_420_w128_8bpc_neon:		16015.7		24969.3
w_mask_422_w4_8bpc_c:	        	858.3		1100.2
w_mask_422_w4_8bpc_neon:		81.5		143.1
w_mask_422_w8_8bpc_c:	        	2447.5		3284.6
w_mask_422_w8_8bpc_neon:		217.5		342.4
w_mask_422_w16_8bpc_c:	        	7673.4		10135.9
w_mask_422_w16_8bpc_neon:		632.5		1062.6
w_mask_422_w32_8bpc_c:	        	28344.9		39090
w_mask_422_w32_8bpc_neon:		2393.4		3963.8
w_mask_422_w64_8bpc_c:	        	68159.6		93447
w_mask_422_w64_8bpc_neon:		6015.7		9928.1
w_mask_422_w128_8bpc_c:	        	169501.2	231702.7
w_mask_422_w128_8bpc_neon:		15847.5		25803.4
w_mask_444_w4_8bpc_c:	        	674.6		862.3
w_mask_444_w4_8bpc_neon:		80.2		135.4
w_mask_444_w8_8bpc_c:	        	2031.4		2693
w_mask_444_w8_8bpc_neon:		209.3		318.7
w_mask_444_w16_8bpc_c:		        6576		8217.4
w_mask_444_w16_8bpc_neon:		627.3		986.2
w_mask_444_w32_8bpc_c:		        26051.7		31593.9
w_mask_444_w32_8bpc_neon:		2374		3671.6
w_mask_444_w64_8bpc_c:		        63600		75849.9
w_mask_444_w64_8bpc_neon:		5957		9335.5
w_mask_444_w128_8bpc_c:		        156964.7	187932.4
w_mask_444_w128_8bpc_neon:		15759.4		24549.5

Limit frame size in pixels to about 16MP, while allowing the fuzzer to test
frame widths and heights above 4096.

Henrik Gramner authored 5 years ago and Henrik Gramner committed 5 years ago

Calling dav1d_get_picture() again after it has already returned with
an error due to a memory allocation failure could result in crashes.

Although doing so is not a proper API usage, and the outcome is going
to be unpredictable, we should at least try to avoid crashing.

Henrik Gramner authored 5 years ago and Henrik Gramner committed 5 years ago

dav1d_submit_frame() could erroneously return 0 when tile data memory
allocation failed.

Fixes an assertion failure in dav1d_parse_obus().

Henrik Gramner authored 5 years ago and Henrik Gramner committed 5 years ago

In the (very unlikely) scenario of a pthread mutex/cond init failure
in the tile state reallocation code some newly allocated mutexes/conds
could leak.

B Krishnan Iyer authored 5 years ago and Martin Storsjö committed 5 years ago

	                     A73             A53
	                Earlier	Now	Earlier	Now

blend_v_w2_8bpc_neon:	122.1	121.3	195.5	195.5
blend_v_w4_8bpc_neon:	248.2	247.5	375.6	358.5
blend_v_w8_8bpc_neon:	210.3	205.2	375.6	358.5
blend_v_w16_8bpc_neon:	252.7	237.1	579.2	590.5
blend_v_w32_8bpc_neon:	347	345.8	997.4	994.1

Henrik Gramner authored 5 years ago and Henrik Gramner committed 5 years ago

Avoid allocating significantly more memory than what is actually used.

Henrik Gramner authored 5 years ago and Henrik Gramner committed 5 years ago

Also eliminate some pointer chasing by allocating tile context buffers
as part of the struct instead of having the struct contain pointers to
separately allocated buffers.

'build_' prefix is reserved by meson, this will become an error in the
future, as indicated by a warning when configuring the build dir.

Closes #285.

For the cdef_filter tests, one could also extend the buffer to
contain 16*11 pixels, to simplify printing it as one rectangular
section.

Extend the common hex_dump function to allow dumping to an arbitrary
FILE* pointer, to reuse it for printing the source pixel buffer in
case of errors.

Martin Storsjö authored 5 years ago and Jean-Baptiste Kempf committed 5 years ago

The speedup for most non-dc-only dct functions is around 9-12x
over the C code generated by GCC 7.3.

Relative speedups vs C for a few functions:

                                              Cortex A53    A72    A73
inv_txfm_add_4x4_dct_dct_0_8bpc_neon:               3.90   4.16   5.65
inv_txfm_add_4x4_dct_dct_1_8bpc_neon:               7.20   8.05  11.19
inv_txfm_add_8x8_dct_dct_0_8bpc_neon:               5.09   6.73   6.45
inv_txfm_add_8x8_dct_dct_1_8bpc_neon:              12.18  10.80  13.05
inv_txfm_add_16x16_dct_dct_0_8bpc_neon:             7.31   9.35  11.17
inv_txfm_add_16x16_dct_dct_1_8bpc_neon:            14.36  13.06  15.93
inv_txfm_add_16x16_dct_dct_2_8bpc_neon:            11.00  10.09  12.05
inv_txfm_add_32x32_dct_dct_0_8bpc_neon:             4.41   5.40   5.77
inv_txfm_add_32x32_dct_dct_1_8bpc_neon:            13.84  13.81  18.04
inv_txfm_add_32x32_dct_dct_2_8bpc_neon:            11.75  11.87  15.22
inv_txfm_add_32x32_dct_dct_3_8bpc_neon:            10.20  10.40  13.13
inv_txfm_add_32x32_dct_dct_4_8bpc_neon:             9.01   9.21  11.56
inv_txfm_add_64x64_dct_dct_0_8bpc_neon:             3.84   4.82   5.28
inv_txfm_add_64x64_dct_dct_1_8bpc_neon:            14.40  12.69  16.71
inv_txfm_add_64x64_dct_dct_4_8bpc_neon:            10.91   9.63  12.67

Some of the specialcased identity_identity transforms for 32x32
give insane speedups over the generic C code:

inv_txfm_add_32x32_identity_identity_0_8bpc_neon: 225.26 238.11 247.07
inv_txfm_add_32x32_identity_identity_1_8bpc_neon: 225.33 238.53 247.69
inv_txfm_add_32x32_identity_identity_2_8bpc_neon:  59.60  61.94  64.63
inv_txfm_add_32x32_identity_identity_3_8bpc_neon:  26.98  27.99  29.21
inv_txfm_add_32x32_identity_identity_4_8bpc_neon:  15.08  15.93  16.56

This keeps the put/prep functions close to the 8tap/bilin functions
that use them.

clock_gettime() is only available since MacOS X 10.12 (Sierra).