Issue 1819 in webm: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset

[w… via monorail]

Status: Available
Owner: ----
CC: jz...@google.com, yunqi...@google.com
Labels: Type-Bug Pri-2

New issue 1819 by w...@google.com: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819

Yunqing and I both saw this error when we run the following command locally:

sh ../libvpx/test/vp9_c_vs_simd_encode.sh

Here are the full error messages:

Files /tmp/vpx_test_4105779/Out-generic-gnu-y4m_360p_10bit_input-rt-1600kbps-cpu0.ivf and /tmp/vpx_test_4105779/Out-armv8-linux-gcc-y4m_360p_10bit_input-rt-1600kbps-cpu0.ivf differ
C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
vp9 test for arm: Done, test failed.

--
You received this message because:
1. The project was configured to send all issue notifications to this address

You may adjust your notification preferences at:
https://bugs.chromium.org/hosting/settings

[Git Watcher via monorail]

Comment #1 on issue 1819 by Git Watcher: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819#c1

The following revision refers to this bug:
https://chromium.googlesource.com/webm/libvpx/+/58a854fb271f99a15f3d2687587f74a6d12c8512

commit 58a854fb271f99a15f3d2687587f74a6d12c8512
Author: Wan-Teh Chang <w...@google.com>
Date: Wed Sep 27 23:12:12 2023

Skip the y4m_360p_10bit_input clip for armv8

It is a known mismatch.

Bug: webm:1819
Change-Id: Ieb707a6f53ddf6c7b0d1202c6520599d3e45da76

[modify] https://crrev.com/58a854fb271f99a15f3d2687587f74a6d12c8512/test/vp9_c_vs_simd_encode.sh

[w… via monorail]

Comment #2 on issue 1819 by w...@google.com: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819#c2

Steps to reproduce:

1. The generic-gnu build:

../libvpx/configure --target=generic-gnu --disable-install-docs --enable-unit-tests --enable-debug --enable-postproc --enable-vp9-postproc --enable-vp9-temporal-denoising --enable-vp9-highbitdepth

make -j

./vpxenc ~/vpx-test-data/crowd_run_360p_10_150f.y4m --bit-depth=10 --profile=2 --codec=vp9 --rt --test-decode=fatal --ivf --threads=1 --static-thresh=0 --tile-columns=0 --tile-rows=0 --end-usage=cbr --kf-max-dist=90000 --lag-in-frames=0 --max-q=58 --min-q=2 --passes=1 --undershoot-pct=50 --overshoot-pct=50 --verbose --row-mt=0 --buf-sz=1000 --buf-initial-sz=500 --buf-optimal-sz=600 --max-intra-rate=300 --resize-allowed=0 --noise-sensitivity=0 --aq-mode=3 --error-resilient=0 --limit=20 --cpu-used=0 --target-bitrate=1600 -o out.ivf

2. The arm64 build:

I did this on a MacBook Pro M1. You can adjust the configure options for your Arm platform.

../libvpx/configure --extra-cflags=-march=armv8.4-a --extra-cxxflags=-march=armv8.4-a --disable-install-docs --enable-unit-tests --enable-debug --enable-postproc --enable-vp9-postproc --enable-vp9-temporal-denoising --enable-vp9-highbitdepth --target=arm64-darwin22-gcc

make -j

./vpxenc ~/vpx-test-data/crowd_run_360p_10_150f.y4m --bit-depth=10 --profile=2 --codec=vp9 --rt --test-decode=fatal --ivf --threads=1 --static-thresh=0 --tile-columns=0 --tile-rows=0 --end-usage=cbr --kf-max-dist=90000 --lag-in-frames=0 --max-q=58 --min-q=2 --passes=1 --undershoot-pct=50 --overshoot-pct=50 --verbose --row-mt=0 --buf-sz=1000 --buf-initial-sz=500 --buf-optimal-sz=600 --max-intra-rate=300 --resize-allowed=0 --noise-sensitivity=0 --aq-mode=3 --error-resilient=0 --limit=20 --cpu-used=0 --target-bitrate=1600 -o out.ivf

3. The two out.ivf files should be the same. But they are different.

[w… via monorail]

Updates:
Cc: mar...@google.com ji...@google.com
Owner: jonat...@arm.com
Status: Assigned

Comment #3 on issue 1819 by w...@google.com: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819#c3

git bisect says:

1a1f50a89d2ba6890024464742bf5a01e034fb45 is the first bad commit
commit 1a1f50a89d2ba6890024464742bf5a01e034fb45
Author: Jonathan Wright <jonatha...@arm.com>
Date: Sat Aug 19 23:41:09 2023 +0100

Use run-time feature detection for Neon DotProd HBD MSE

Arm Neon DotProd implementations of vpx_highbd_8_mse<w>x<h> currently
need to be enabled at compile time since they're guarded by #ifdef
feature macros. Now that run-time feature detection has been enabled
for Arm platforms, expose these implementations with distinct
*neon_dotprod names in a separate file and wire them up to the build
system and rtcd.pl. Also add new test cases for the new functions.

Change-Id: I26be6fb587258c8fa9fbf03509b7602358a001a8

test/variance_test.cc | 10 ++++
vpx_dsp/arm/highbd_variance_neon.c | 67 +--------------------
vpx_dsp/arm/highbd_variance_neon_dotprod.c | 96 ++++++++++++++++++++++++++++++
vpx_dsp/vpx_dsp.mk | 1 +
vpx_dsp/vpx_dsp_rtcd_defs.pl | 8 +--
5 files changed, 113 insertions(+), 69 deletions(-)
create mode 100644 vpx_dsp/arm/highbd_variance_neon_dotprod.c

[w… via monorail]

Comment #4 on issue 1819 by w...@google.com: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819#c4

Jonathan,

My debugging showed that vpx_highbd_8_mse16x16() is called for any bit depth, not just 8. See psnr.c:147:

127 static int64_t highbd_get_sse(const uint8_t *a, int a_stride, const uint8_t *b,
128 int b_stride, int width, int height) {
129 int64_t total_sse = 0;
130 int x, y;
...
142 for (y = 0; y < height / 16; ++y) {
143 const uint8_t *pa = a;
144 const uint8_t *pb = b;
145 unsigned int sse;
146 for (x = 0; x < width / 16; ++x) {
147 vpx_highbd_8_mse16x16(pa, a_stride, pb, b_stride, &sse);
148 total_sse += sse;
149 pa += 16;
150 pb += 16;
151 }
152 a += 16 * a_stride;
153 b += 16 * b_stride;
154 }
155 return total_sse;
156 }

So we can't use vdotq_u32, which operates on uint8x16_t. I am afraid that we will need to delete highbd_variance_neon_dotprod.c.

[jonat… via monorail]

Comment #5 on issue 1819 by jonat...@arm.com: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819#c5

Hi @wtc, we fixed this problem in libaom with the following change: https://aomedia-review.googlesource.com/c/aom/+/175063

I think porting that fix across - or something to that effect - would be the correct solution here.

[w… via monorail]

Updates:
Cc: jonat...@arm.com
Owner: w...@google.com
Status: Started

Comment #6 on issue 1819 by w...@google.com: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819#c6

Hi Jonathan,

Thank you very much for the info. I agree with your suggestion. If you or your colleague has started working on it, please let me know. Otherwise I can take a stab at porting your fix to libvpx today.

[jonat… via monorail]

Comment #7 on issue 1819 by jonat...@arm.com: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819#c7

Nobody has started a fix on our side, so do go ahead - thanks!

[w… via monorail]

Comment #8 on issue 1819 by w...@google.com: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819#c8

This requires porting libaom's aom_dsp/sse.c and relates files.

Git history shows that the authors of these files are Google employees and contractors and Arm employees. So we (the two companies) can recontribute our code to libvpx.

[Git Watcher via monorail]

Comment #9 on issue 1819 by Git Watcher: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819#c9

The following revision refers to this bug:

https://chromium.googlesource.com/webm/libvpx/+/6512f994da13e2f27e6a7bd449efee0a374b55b7

commit 6512f994da13e2f27e6a7bd449efee0a374b55b7
Author: Wan-Teh Chang <w...@google.com>
Date: Fri Sep 29 23:58:26 2023

Disable vpx_highbd_8_mse16x16_neon_dotprod, etc.

These functions assume the uint16_t samples are <= 255 (bit depth 8),
but vpx_highbd_8_mse16x16() is called for any bit depth, not just 8.

A better fix is to port the libaom CL
https://aomedia-review.googlesource.com/c/aom/+/175063 to libvpx, but
that requires porting aom_sse() and aom_highbd_sse() to libvpx, which is
quite involved. So disable vpx_highbd_8_mse16x16_neon_dotprod, etc.
first.

Bug: webm:1819
Change-Id: If495a5dedc58d9981317b9993c9fbb81ff3ab50c

[modify] https://crrev.com/6512f994da13e2f27e6a7bd449efee0a374b55b7/test/variance_test.cc
[modify] https://crrev.com/6512f994da13e2f27e6a7bd449efee0a374b55b7/test/vp9_c_vs_simd_encode.sh
[modify] https://crrev.com/6512f994da13e2f27e6a7bd449efee0a374b55b7/vpx_dsp/vpx_dsp.mk
[modify] https://crrev.com/6512f994da13e2f27e6a7bd449efee0a374b55b7/vpx_dsp/vpx_dsp_rtcd_defs.pl

[w… via monorail]

Comment #10 on issue 1819 by w...@google.com: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819#c10

Hi Jonathan,

One reason I decided to disable the vpx_highbd_8_mse16x16_neon_dotprod, etc. functions is that they are also used for all bit depths via the highbd_get_block_variance_fn(bsize, 8) call in vp9/encoder/vp9_firstpass.c. That call won't be fixed by porting your libaom CL https://aomedia-review.googlesource.com/c/aom/+/175063 to libvpx.

In libaom, highbd_get_block_variance_fn() is not called with 8 as the second argument.

I am wondering if the highbd_get_block_variance_fn(bsize, 8) call can be replaced by vpx_highbd_sse() (when it is available).

[jonat… via monorail]

Comment #11 on issue 1819 by jonat...@arm.com: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819#c11

Thanks for the update.

FYI, I have just written a patch[1] to optimize aom_highbd_sse_neon: 2-4x the perf of the old implementation. You may wish to hold off porting to libvpx until that lands.

[1] https://aomedia-review.googlesource.com/c/aom/+/181967

[yunqi… via monorail]

Updates:
Cc: deepa...@ittiam.com

Comment #13 on issue 1819 by yunqi...@google.com: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819#c13

Issue 1811 has been merged into this issue.

[Git Watcher via monorail]

Comment #14 on issue 1819 by Git Watcher: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819#c14

The following revision refers to this bug:

https://chromium.googlesource.com/webm/libvpx/+/a8db542b24a50060e996feda00b0538c7c334909

commit a8db542b24a50060e996feda00b0538c7c334909
Author: Wan-Teh Chang <w...@google.com>
Date: Fri Sep 29 17:45:32 2023

Add vpx_sse and vpx_highbd_sse

The code is ported from libaom's aom_sse and aom_highbd_sse at
commit 1e20d2da96515524864b21010dbe23809cff2e9b.

The vpx_sse and vpx_highbd_sse functions will be used by vpx_dsp/psnr.c.

Bug: webm:1819
Change-Id: I4fbffa9000ab92755de5387b1ddd4370cb7020f7

[add] https://crrev.com/a8db542b24a50060e996feda00b0538c7c334909/vpx_dsp/arm/sse_neon_dotprod.c
[modify] https://crrev.com/a8db542b24a50060e996feda00b0538c7c334909/test/sum_squares_test.cc
[add] https://crrev.com/a8db542b24a50060e996feda00b0538c7c334909/vpx_dsp/x86/sse_sse4.c
[add] https://crrev.com/a8db542b24a50060e996feda00b0538c7c334909/vpx_dsp/sse.c
[add] https://crrev.com/a8db542b24a50060e996feda00b0538c7c334909/vpx_dsp/arm/highbd_sse_neon.c
[add] https://crrev.com/a8db542b24a50060e996feda00b0538c7c334909/vpx_dsp/x86/sse_avx2.c
[modify] https://crrev.com/a8db542b24a50060e996feda00b0538c7c334909/vpx_dsp/arm/sum_neon.h
[add] https://crrev.com/a8db542b24a50060e996feda00b0538c7c334909/vpx_dsp/arm/sse_neon.c
[modify] https://crrev.com/a8db542b24a50060e996feda00b0538c7c334909/vpx_dsp/vpx_dsp.mk
[modify] https://crrev.com/a8db542b24a50060e996feda00b0538c7c334909/vpx_dsp/vpx_dsp_rtcd_defs.pl

[Git Watcher via monorail]

Comment #15 on issue 1819 by Git Watcher: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819#c15

The following revision refers to this bug:

https://chromium.googlesource.com/webm/libvpx/+/7dfe343199381bddddc5eaa648e947876979b61b

commit 7dfe343199381bddddc5eaa648e947876979b61b
Author: Wan-Teh Chang <w...@google.com>
Date: Thu Sep 28 16:26:58 2023

Use vpx_sse instead of vpx_mse to compute SSE

Use vpx_sse and vpx_highbd_sse instead of vpx_mse16x16 and
vpx_highbd_8_mse16x16 respectively to compute SSE for PSNR
calculations. This solves an issue whereby vpx_highbd_8_mse16x16
was being used to calculate SSE for 10- and 12-bit input.

This is a port of the libaom CL
https://aomedia-review.googlesource.com/c/aom/+/175063
by Jonathan Wright <jonatha...@arm.com>.

Bug: webm:1819
Change-Id: I37e3ac72835e67ccb44ac89a4ed16df62c2169a7

[modify] https://crrev.com/7dfe343199381bddddc5eaa648e947876979b61b/vpx_dsp/psnr.c

[Git Watcher via monorail]

Comment #16 on issue 1819 by Git Watcher: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819#c16

The following revision refers to this bug:

https://chromium.googlesource.com/webm/libvpx/+/c4c92080545970899488ab27944792a95c7131a2

commit c4c92080545970899488ab27944792a95c7131a2
Author: Wan-Teh Chang <w...@google.com>
Date: Tue Dec 05 22:29:37 2023

Remove SSE code for 128x* blocks

The maximum block size is 64x64 in VP9.

Bug: webm:1819
Change-Id: If9802be9f81b51dbcdbc8a68d5afe48ca6d3d0e7

[modify] https://crrev.com/c4c92080545970899488ab27944792a95c7131a2/vpx_dsp/arm/sse_neon_dotprod.c
[modify] https://crrev.com/c4c92080545970899488ab27944792a95c7131a2/vpx_dsp/x86/sse_sse4.c
[modify] https://crrev.com/c4c92080545970899488ab27944792a95c7131a2/vpx_dsp/arm/highbd_sse_neon.c
[modify] https://crrev.com/c4c92080545970899488ab27944792a95c7131a2/vpx_dsp/x86/sse_avx2.c
[modify] https://crrev.com/c4c92080545970899488ab27944792a95c7131a2/vpx_dsp/arm/sse_neon.c

[Git Watcher via monorail]

Updates:
Labels: merge-merged-venetian

Comment #17 on issue 1819 by Git Watcher: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819#c17

The following revision refers to this bug:

https://chromium.googlesource.com/webm/libvpx/+/0d5811e4effb62672ec60aed5bd21553af158791

commit 0d5811e4effb62672ec60aed5bd21553af158791
Author: Wan-Teh Chang <w...@google.com>

Date: Thu Sep 28 16:26:58 2023

Use vpx_sse instead of vpx_mse to compute SSE

Use vpx_sse and vpx_highbd_sse instead of vpx_mse16x16 and
vpx_highbd_8_mse16x16 respectively to compute SSE for PSNR
calculations. This solves an issue whereby vpx_highbd_8_mse16x16
was being used to calculate SSE for 10- and 12-bit input.

This is a port of the libaom CL
https://aomedia-review.googlesource.com/c/aom/+/175063
by Jonathan Wright <jonatha...@arm.com>.

Bug: webm:1819
Change-Id: I37e3ac72835e67ccb44ac89a4ed16df62c2169a7

(cherry picked from commit 7dfe343199381bddddc5eaa648e947876979b61b)

[modify] https://crrev.com/0d5811e4effb62672ec60aed5bd21553af158791/vpx_dsp/psnr.c

[Git Watcher via monorail]

Comment #18 on issue 1819 by Git Watcher: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819#c18

The following revision refers to this bug:

https://chromium.googlesource.com/webm/libvpx/+/6fca4de48eaa110ada3ca849f445ca3754fb869a

commit 6fca4de48eaa110ada3ca849f445ca3754fb869a
Author: Wan-Teh Chang <w...@google.com>

Date: Tue Dec 05 22:29:37 2023

Remove SSE code for 128x* blocks

The maximum block size is 64x64 in VP9.

Bug: webm:1819
Change-Id: If9802be9f81b51dbcdbc8a68d5afe48ca6d3d0e7

(cherry picked from commit c4c92080545970899488ab27944792a95c7131a2)

[modify] https://crrev.com/6fca4de48eaa110ada3ca849f445ca3754fb869a/vpx_dsp/arm/sse_neon_dotprod.c
[modify] https://crrev.com/6fca4de48eaa110ada3ca849f445ca3754fb869a/vpx_dsp/x86/sse_sse4.c
[modify] https://crrev.com/6fca4de48eaa110ada3ca849f445ca3754fb869a/vpx_dsp/arm/highbd_sse_neon.c
[modify] https://crrev.com/6fca4de48eaa110ada3ca849f445ca3754fb869a/vpx_dsp/x86/sse_avx2.c
[modify] https://crrev.com/6fca4de48eaa110ada3ca849f445ca3754fb869a/vpx_dsp/arm/sse_neon.c

[Git Watcher via monorail]

Comment #19 on issue 1819 by Git Watcher: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819#c19

The following revision refers to this bug:

https://chromium.googlesource.com/webm/libvpx/+/41e0655e5e612726633cc38d92c0257f9f0dc2dd

commit 41e0655e5e612726633cc38d92c0257f9f0dc2dd
Author: Gerda Zsejke More <gerdazse...@arm.com>
Date: Tue Jan 16 15:13:50 2024

Fix highbd_get_block_variance_fn input parameter

Instead of always calling highbd_get_block_variance_fn with bit depth
8 use the macroblock's bit depth.

Bug: webm:1819
Change-Id: Ib4b19703384e897ee9ffeef73a11a8af2d262558

[modify] https://crrev.com/41e0655e5e612726633cc38d92c0257f9f0dc2dd/vp9/encoder/vp9_firstpass.c

[Git Watcher via monorail]

Comment #20 on issue 1819 by Git Watcher: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819#c20

The following revision refers to this bug:

https://chromium.googlesource.com/webm/libvpx/+/e001eeb5bcdfefba2f6e85b7307e410383af5d57

commit e001eeb5bcdfefba2f6e85b7307e410383af5d57

Author: Gerda Zsejke More <gerdazse...@arm.com>

Date: Tue Jan 16 15:21:42 2024

Enable Neon Dotprod impl for HBD MSE

Enable Neon Dotprod 8-bit high bitdepth implementation for MSE
function as it is now not called with bit depth 10 or 12.

Bug: webm:1819
Change-Id: I9d1d506401aa0523fba2d8ea4978dc00fdacbb95

[modify] https://crrev.com/e001eeb5bcdfefba2f6e85b7307e410383af5d57/test/variance_test.cc
[modify] https://crrev.com/e001eeb5bcdfefba2f6e85b7307e410383af5d57/vpx_dsp/vpx_dsp.mk
[modify] https://crrev.com/e001eeb5bcdfefba2f6e85b7307e410383af5d57/vpx_dsp/vpx_dsp_rtcd_defs.pl

[jonat… via monorail]

Updates:
Status: Fixed

Comment #21 on issue 1819 by jonat...@arm.com: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819#c21

Verified that Gerda's recent patches fix the issue - at least in the reproducer test case in comment #2.

Marking as fixed but deferring to wtc@ to close.

[w… via monorail]

Updates:
Status: Verified

Comment #22 on issue 1819 by w...@google.com: vp9_c_vs_simd_encode.sh: C vs armv8-linux-gcc encode mismatches for y4m_360p_10bit_input, at 1600 kbps, speed 0, rt preset
https://bugs.chromium.org/p/webm/issues/detail?id=1819#c22

Gerda and Jonathan: Thank you!