From: softworkz
Signed-off-by: softworkz
---
fftools/resources/resman.c | 7 +++
fftools/resources/resman.h | 10 +-
2 files changed, 4 insertions(+), 13 deletions(-)
diff --git a/fftools/resources/resman.c b/fftools/resources/resman.c
index a9e21626fa..f098bd6687 100644
--- a/fft
> -Original Message-
> From: ffmpeg-devel On Behalf Of softworkz .
> Sent: Samstag, 31. Mai 2025 04:46
> To: FFmpeg development discussions and patches
> Cc: Kieran Kunhya
> Subject: Re: [FFmpeg-devel] [PATCH 2/2] fftools/resman: use inflate loop with
> increasing buffer size
>
>
>
From: softworkz
Signed-off-by: softworkz
---
fftools/resources/resman.c | 51 +++---
1 file changed, 36 insertions(+), 15 deletions(-)
diff --git a/fftools/resources/resman.c b/fftools/resources/resman.c
index f098bd6687..e3e082abbf 100644
--- a/fftools/resource
and other cosmetic fixes.
As per review by Ramiro Polla in message: "Re: [FFmpeg-devel]
[FFmpeg-cvslog] fftools/resources: Add resource manager files with
build-time compression"
V2
* Use two-pass decompression to return resources in actual-size memory
buffers
(in response to comment by K
> -Original Message-
> From: ffmpeg-devel On Behalf Of Kieran
> Kunhya via ffmpeg-devel
> Sent: Samstag, 31. Mai 2025 04:39
> To: FFmpeg development discussions and patches
> Cc: Kieran Kunhya
> Subject: Re: [FFmpeg-devel] [PATCH 2/2] fftools/resman: use inflate loop with
> increasing
>
> }
> +
> +if (strm.avail_out == 0) {
> +chunk *= 8;
>
*8 seems high
+uint8_t *tmp_buf = av_realloc(buf, chunk + 1);
> +if (!tmp_buf) {
> +inflateEnd(&strm);
> +av_free(buf);
> +return AVERROR(E
> -Original Message-
> From: ffmpeg-devel On Behalf Of James Almer
> Sent: Donnerstag, 29. Mai 2025 05:07
> To: ffmpeg-devel@ffmpeg.org
> Subject: [FFmpeg-devel] [PATCH 1/2] fftools/ffmpeg_filter: make
> FilterGraphPriv private again
>
> As the name implies, it's a struct meant to be i
On Sat, May 24, 2025 at 10:02 PM Nuo Mi wrote:
>
>
> On Fri, May 9, 2025 at 2:37 AM Wu Jianhua wrote:
>
>> Andreas Rheinhardt:
>> > toq...@gmail.com:
>> >> From: Wu Jianhua
>> >>
>> >> Signed-off-by: Wu Jianhua
>> >> ---
>> >> libavcodec/h274.c | 163 ++
> -Original Message-
> From: ffmpeg-devel On Behalf Of Ramiro Polla
> Sent: Freitag, 30. Mai 2025 13:24
> To: ffmpeg-devel@ffmpeg.org
> Subject: Re: [FFmpeg-devel] [FFmpeg-cvslog] fftools/resources: Add resource
> manager files with build-time compression
>
> On Thu, May 15, 2025 at 11:
On Tue, May 27, 2025 at 10:52 PM Gabriel Hege wrote:
> On 25.05.25 04:08, Nuo Mi wrote:
> > On Thu, May 22, 2025 at 4:35 PM Gabriel Hege wrote:
> >
> >>
> >> On 17.05.25 04:26, Nuo Mi wrote:
> >>> On Fri, May 16, 2025 at 7:05 PM Gabriel Hege wrote:
> >>>
> This is a resubmission with a cor
From: softworkz
Signed-off-by: softworkz
---
fftools/resources/resman.c | 38 --
1 file changed, 24 insertions(+), 14 deletions(-)
diff --git a/fftools/resources/resman.c b/fftools/resources/resman.c
index f098bd6687..0dae272c89 100644
--- a/fftools/resource
From: softworkz
Signed-off-by: softworkz
---
fftools/resources/resman.c | 7 +++
fftools/resources/resman.h | 10 +-
2 files changed, 4 insertions(+), 13 deletions(-)
diff --git a/fftools/resources/resman.c b/fftools/resources/resman.c
index a9e21626fa..f098bd6687 100644
--- a/fft
and other cosmetic fixes.
As per review by Ramiro Polla in message: "Re: [FFmpeg-devel]
[FFmpeg-cvslog] fftools/resources: Add resource manager files with
build-time compression"
softworkz (2):
fftools/resman: remove unused includes, fix declaration
fftools/resman: use inflate loop with incre
Hello James,
> -Original Message-
> From: ffmpeg-devel On Behalf Of James Almer
> Sent: Donnerstag, 29. Mai 2025 05:07
> To: ffmpeg-devel@ffmpeg.org
> Subject: [FFmpeg-devel] [PATCH 2/2] fftools/ffmpeg_filter: make
> InputFilterPriv and OutputFilterPriv private again
>
> As the names imp
Patch attached.
- Andreas
From 1c4456776072551b303484083da4fb1c7e1a44a2 Mon Sep 17 00:00:00 2001
From: Andreas Rheinhardt
Date: Sat, 31 May 2025 03:32:47 +0200
Subject: [PATCH] tests/fate-run: Remove intermediate files from enc-external
tests
Signed-off-by: Andreas Rheinhardt
---
tests/fate-r
Hi
On Sat, May 31, 2025 at 02:45:30AM +0200, Andreas Rheinhardt wrote:
> Michael Niedermayer:
> > non flat extradata is problematic and was missed by reviewers
> >
> > Found-by: mkver and jamrial
>
> Actual patch LGTM, but I'd rather prefer to have my actual name here.
will apply with your full
On Fri, May 30, 2025 at 12:07 AM Frank Plowman
wrote:
> Signed-off-by: Frank Plowman
> ---
> Was requested on IRC.
> ---
> MAINTAINERS | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/MAINTAINERS b/MAINTAINERS
> index d1d87752b9..485cca6c75 100644
> --- a/MAINTAINERS
> +
On 5/30/2025 10:17 PM, Michael Niedermayer wrote:
Hi
On Sat, May 31, 2025 at 02:45:30AM +0200, Andreas Rheinhardt wrote:
Michael Niedermayer:
non flat extradata is problematic and was missed by reviewers
Found-by: mkver and jamrial
I did not find or report this issue, it was Andreas. I just
> -Original Message-
> From: ffmpeg-devel On Behalf Of James Almer
> Sent: Donnerstag, 29. Mai 2025 05:07
> To: ffmpeg-devel@ffmpeg.org
> Subject: [FFmpeg-devel] [PATCH 2/2] fftools/ffmpeg_filter: make
> InputFilterPriv and OutputFilterPriv private again
>
> As the names imply, they ar
Michael Niedermayer:
> non flat extradata is problematic and was missed by reviewers
>
> Found-by: mkver and jamrial
Actual patch LGTM, but I'd rather prefer to have my actual name here.
> This reverts commit 574f634e49847e2225ee50013afebf0de03ef013.
> ---
> libavcodec/vorbis_parser.h
Romain Beauxis:
> ---
> libavcodec/vorbis_parser.h | 11
> libavcodec/vorbisdec.c | 75 +-
> libavformat/oggparsevorbis.c | 67 ++-
> tests/ref/fate/ogg-vorbis-chained-meta.txt | 3 -
> tests/ref/fate/trac-
non flat extradata is problematic and was missed by reviewers
Found-by: mkver and jamrial
This reverts commit 574f634e49847e2225ee50013afebf0de03ef013.
---
libavcodec/vorbis_parser.h | 11
libavcodec/vorbisdec.c | 75 +-
libavformat/ogg
Offers a modest performance gain due to the switch from naive linear
probling to robin hood.
Signed-off-by: Emma Worley
---
libavcodec/dxv.h| 2 +-
libavcodec/dxvenc.c | 123
2 files changed, 35 insertions(+), 90 deletions(-)
diff --git a/libav
Improves compatibility with Resolume products by adding an additional
hashtable for DXT color+LUT combinations, and padding the DXT texture
dimensions to the next largest multiple of 16. Produces identical
packets to Resolume Alley in manual tests.
Signed-off-by: Emma Worley
---
libavcodec/dxven
Adds a generic hash table with the DXV encoder as an initial use case.
Signed-off-by: Emma Worley
---
libavcodec/Makefile | 2 +
libavcodec/hashtable.c | 214 +++
libavcodec/hashtable.h | 94 +++
libavcodec/tests/hashtable.c | 1
> -Original Message-
> From: ffmpeg-devel On Behalf Of Ramiro Polla
> Sent: Freitag, 30. Mai 2025 12:52
> To: ffmpeg-devel@ffmpeg.org
> Subject: [FFmpeg-devel] [PATCH v3 1/2] fftools/resources: clean up and improve
> resource manager build system
>
> - move .gitignore entries to main .
Hi Manuel
On Fri, May 30, 2025 at 10:54:01PM +0200, Manuel Lauss wrote:
> Hi Michael,
>
> On Fri, May 30, 2025 at 9:52 PM Michael Niedermayer
> wrote:
> >
> > Hi Manual
> >
> > On Wed, May 14, 2025 at 03:39:53AM +0200, Michael Niedermayer wrote:
> > > k is always 4 here and it seems this is not
On Fri, May 30, 2025 at 10:48:29PM +0200, Andreas Rheinhardt wrote:
> Michael Niedermayer:
> > On Sat, May 17, 2025 at 03:03:19PM +0200, Andreas Rheinhardt wrote:
> >> Andreas Rheinhardt:
> >>> Kacper Michajłow:
> It's not valid to call avio_close() on context that has not been open
> wit
Patch attached.
- Andreas
From b45630a3525fe7d634500e981879a3d07a588855 Mon Sep 17 00:00:00 2001
From: Andreas Rheinhardt
Date: Fri, 30 May 2025 19:11:45 +0200
Subject: [PATCH] avcodec/libaomenc: Avoid code duplication when setting
options
(Some more options could be set via this method
if the
Hi Michael,
On Fri, May 30, 2025 at 9:52 PM Michael Niedermayer
wrote:
>
> Hi Manual
>
> On Wed, May 14, 2025 at 03:39:53AM +0200, Michael Niedermayer wrote:
> > k is always 4 here and it seems this is not what was intended
> > replacing it with 0 works but it may be wrong
> >
> > This needs revi
Michael Niedermayer:
> On Sat, May 17, 2025 at 03:03:19PM +0200, Andreas Rheinhardt wrote:
>> Andreas Rheinhardt:
>>> Kacper Michajłow:
It's not valid to call avio_close() on context that has not been open
with avio_open().
This fixes use of custom IO. (io_open / io_close2 callb
Hi
On Mon, May 26, 2025 at 05:28:50PM -0500, Romain Beauxis wrote:
> Le lun. 26 mai 2025 à 13:38, Michael Niedermayer a
> écrit :
> >
> > On Sat, May 24, 2025 at 01:14:04PM -0500, Romain Beauxis wrote:
> > > Sample available at:
> https://www.dropbox.com/scl/fo/xrtrna2rxr1j354hrtymq/AGwemlxHYecBL
On Fri, May 30, 2025 at 3:52 AM Maryla Ustarroz-Calonge via
ffmpeg-devel wrote:
>
> Signed-off-by: Maryla Ustarroz-Calonge
> ---
> libavcodec/libaomdec.c | 62 ++
> libavcodec/libaomenc.c | 60
> libavcodec/version.
Hi Manual
On Wed, May 14, 2025 at 03:39:53AM +0200, Michael Niedermayer wrote:
> k is always 4 here and it seems this is not what was intended
> replacing it with 0 works but it may be wrong
>
> This needs review
>
> Fixes: out of array read
> Fixes:
> 409593384/clusterfuzz-testcase-minimized-f
On Wed, May 14, 2025 at 03:39:52AM +0200, Michael Niedermayer wrote:
> Fixes: Assertion sc->slice_coding_mode == 0 failed at
> libavcodec/ffv1enc.c:1667
> Fixes:
> 408838118/clusterfuzz-testcase-minimized-ffmpeg_AV_CODEC_ID_FFV1_fuzzer-6493138204295168
>
> Found-by: continuous fuzzing process
>
On Sat, May 17, 2025 at 03:03:19PM +0200, Andreas Rheinhardt wrote:
> Andreas Rheinhardt:
> > Kacper Michajłow:
> >> It's not valid to call avio_close() on context that has not been open
> >> with avio_open().
> >>
> >> This fixes use of custom IO. (io_open / io_close2 callbacks)
> >>
> >> Note tha
Signed-off-by: Emma Worley
---
MAINTAINERS | 1 +
1 file changed, 1 insertion(+)
diff --git a/MAINTAINERS b/MAINTAINERS
index d1d87752b9..0fba390938 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -176,6 +176,7 @@ Codecs:
dss_sp.c Oleksij Rempel
dv.c
On Sat, Apr 26, 2025 at 01:05:07AM +0900, Emma Worley wrote:
> Adds a generic hash table with the DXV encoder as an initial use case.
>
> Signed-off-by: Emma Worley
> ---
> libavcodec/Makefile | 2 +
> libavcodec/hashtable.c | 214 +++
> libavcode
Hi Emma
On Wed, May 28, 2025 at 03:35:29PM -0700, Emma Worley wrote:
> I successfully used this patch series to encode DXV files for a couple
> live events this past weekend and did not encounter any decoding
> issues with Resolume's first-party Arena software. These DXVs included
> inputs that ca
On May 30, 2025 4:10:45 AM PDT, Ramiro Polla wrote:
> I'll postpone this part of the patchset (patches 1 to 3) for a while
> to give more time for people to comment. If anybody can find a system
> where zlib doesn't support gzip, please let us know.
idk if anyone is using it as the system zlib
On 5/30/2025 1:44 PM, Andreas Rheinhardt wrote:
Patches attached.
- Andreas
For #11617, something simpler like the following should be enough
diff --git a/libavcodec/dovi_rpuenc.c b/libavcodec/dovi_rpuenc.c
index 2e1f8be08e..bbc66fe2e8 100644
--- a/libavcodec/dovi_rpuenc.c
+++ b/libavcodec/d
On Fri, 30 May 2025 18:44:33 +0200 Andreas Rheinhardt
wrote:
> Patches attached.
Patches LGTM.
___
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-dev
Patches attached.
- Andreas
From 19b550d20a2f00e776c3e17b56b3ec55abf71752 Mon Sep 17 00:00:00 2001
From: Andreas Rheinhardt
Date: Fri, 30 May 2025 18:06:55 +0200
Subject: [PATCH 1/4] avcodec/dovi_rpuenc: Avoid intermediate codec par in
ff_dovi_configure()
It invalidates (removes by duplicates)
Patches attached.
- Andreas
From a33bb0298c0eaf7816c36be7c5558cfcc5a3f37c Mon Sep 17 00:00:00 2001
From: Andreas Rheinhardt
Date: Fri, 30 May 2025 12:40:16 +0200
Subject: [PATCH 1/4] avcodec/hpeldsp: Remove duplicate pel functions
Signed-off-by: Andreas Rheinhardt
---
libavcodec/hpeldsp.c | 5
Dmitrii Ovchinnikov :
> Wu Jianhuaw
>> Dmitrii Ovchinnikov :
>>> Certain components require ID3D11/ID3D12 textures to be created with
>>> additional BindFlags or MiscFlags. FFmpeg currently provides no
>>> mechanism to configure these flags externally, leaving no way to
>>> satisfy such components
---
libavutil/hwcontext_amf.c | 40 ---
1 file changed, 21 insertions(+), 19 deletions(-)
diff --git a/libavutil/hwcontext_amf.c b/libavutil/hwcontext_amf.c
index 7903109445..a020b0e2e1 100644
--- a/libavutil/hwcontext_amf.c
+++ b/libavutil/hwcontext_amf.c
@@ -
---
libavutil/hwcontext_amf.c | 36
1 file changed, 36 insertions(+)
diff --git a/libavutil/hwcontext_amf.c b/libavutil/hwcontext_amf.c
index 24731c20ec..7903109445 100644
--- a/libavutil/hwcontext_amf.c
+++ b/libavutil/hwcontext_amf.c
@@ -28,6 +28,9 @@
#if C
---
configure | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/configure b/configure
index a1f483f02c..b9de840ec2 100755
--- a/configure
+++ b/configure
@@ -7476,7 +7476,7 @@ fi
enabled amf &&
check_cpp_condition amf "AMF/core/Version.h" \
-"(AMF_VERSION_MAJOR <<
2 patches to fix video playback of Rebel Assault 1 SEGA-CD Version.
The code for this was reversed from the latest DOS EXE and tested
with content from the SEGA-CD release.
See https://ibb.co/QvsbpWY9 for before/after screenshots.
Manuel Lauss (2):
avcodec/sanm: fix codec33/34 tile generator
Maryla Ustarroz-Calonge via ffmpeg-devel:
> The new fate sample av1/metadata_hdr10_plus.ivf used in the second
> test is the output of the first test.
>
> Signed-off-by: Maryla Ustarroz-Calonge
> ---
> tests/Makefile | 1 +
> tests/fate/av1.mak | 11 +
On Fri, 30 May 2025 04:23:12 +0200 Michael Niedermayer
wrote:
> On Tue, May 27, 2025 at 09:55:33AM +0200, Niklas Haas wrote:
> > From: Niklas Haas
> >
> > This covers most 8-bit and 16-bit ops, and some 32-bit ops. It also covers
> > all
> > floating point operations. While this is not yet 100%
On Thu, May 15, 2025 at 11:11 PM softworkz wrote:
>
> ffmpeg | branch: master | softworkz | Thu May 15
> 23:07:54 2025 +0200| [517a8055655798970d94a4c5ea912511362520ea] | committer:
> softworkz
>
> fftools/resources: Add resource manager files with build-time compression
>
> Compression require
codec31/32 are like codec1/3 (RLE coding) but with 2 4-bit pixels per byte.
---
libavcodec/sanm.c | 59 +++
1 file changed, 59 insertions(+)
diff --git a/libavcodec/sanm.c b/libavcodec/sanm.c
index 46cad3c703..975f33992f 100644
--- a/libavcodec/sanm.c
+
tested with LVL11/L11PLAY.ANM from Rebel Assault Sega-CD release.
---
libavcodec/sanm.c | 33 +++--
1 file changed, 15 insertions(+), 18 deletions(-)
diff --git a/libavcodec/sanm.c b/libavcodec/sanm.c
index 00bfef00fe..46cad3c703 100644
--- a/libavcodec/sanm.c
+++ b/li
Hi Timo,
On Tue, May 27, 2025 at 7:55 PM Ramiro Polla wrote:
> On Tue, May 27, 2025 at 3:19 PM Timo Rothenpieler
> wrote:
> > On 27.05.2025 03:33, Ramiro Polla wrote:
> > > We currently test at build-time whether zlib supports decoding gzip.
> > > This is not needed for the build to succeed, si
The new fate sample av1/metadata_hdr10_plus.ivf used in the second
test is the output of the first test.
Signed-off-by: Maryla Ustarroz-Calonge
---
tests/Makefile | 1 +
tests/fate/av1.mak | 11 ++
tests/ref/fate/libaom-hdr10-plus | 52 +
Signed-off-by: Maryla Ustarroz-Calonge
---
libavcodec/libaomdec.c | 62 ++
libavcodec/libaomenc.c | 60
libavcodec/version.h | 2 +-
3 files changed, 123 insertions(+), 1 deletion(-)
diff --git a/libavcodec/libao
On Tue, May 27, 2025 at 10:13 PM softworkz .
wrote:
> > -Original Message-
> > From: ffmpeg-devel On Behalf Of
> > Ramiro Polla
> > Sent: Dienstag, 27. Mai 2025 21:59
> > To: FFmpeg development discussions and patches > de...@ffmpeg.org>
> > Subject: Re: [FFmpeg-devel] [PATCH 4/5] ffbuil
---
fftools/Makefile | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/fftools/Makefile b/fftools/Makefile
index b6f3438e11..97da49b476 100644
--- a/fftools/Makefile
+++ b/fftools/Makefile
@@ -93,4 +93,4 @@ uninstall-progs:
$(RM) $(addprefix "$(BINDIR)/", $(ALLAVPROGS))
- move .gitignore entries to main .gitignore;
- move vpath directives to main Makefile;
- remove superfluous comments;
- turn css minification sed command into a one-liner;
- deduplicate targets depending on CONFIG_RESOURCE_COMPRESSION;
- introduce common .res pattern for resource files;
- remove R
Hi Jianhua,
The patch is designed for cases when a component doesn't allocate frames,
but only uses frames that come from outside (for example, from a decoder)
and has requirements for them. In cases where flags are configured this
way, we can avoid unnecessary surface copying, while the default b
On Fri, 30 May 2025, Dmitriy Kovalenko wrote:
=== Feedback response ===
FWIW, the procedure is to respond to inline comments by replying to the
mails where those comments were made. When they're included here, they end
up as part of your suggested commit message.
Anyway, now this time, the
On Fri, 30 May 2025, Dmitriy Kovalenko wrote:
If you with "non-performant mobile" mean small in-order cores, most of them can handle repeated
accumulation like these even faster, if you sequence these so that all accumulations to one register is
sequentially. E.g. first all "smlal \u_dst1\().4
I'm sorry for the previous patch it seems to be something happening off with
the corrupted
patch got sent at the outlook step, I'll keep using send-email.
=== __every single__ inline comment response ===
> This is an unrelated change
Fixed and resolved
> The patch adds trailing whitespace here
=== Feedback response ===
> Also, with that fixed, this fails to properly back up and restore registers
> v8-v15; checkasm doesn't notice this on macOS, but on Linux and windows,
> checkasm has a call wrapper which does detect such issues.
I managed to rewrite the function to avoid using any ca
Hi,
Thanks for your review and inputs. I understand.
Instead of making the changes here, I'm trying to create a hw_frames_ctx with
required config and pass it while opening the decoder, but I'm facing
exceptions at av_buffer_unref(&dst->hw_frames_ctx) in
update_context_from_thread. Call stack
From: daichengrong
On Banana PI F3:
hevc_idct_32x32_8_c:118807.4 ( 1.00x)
hevc_idct_32x32_8_rvv_i64: 13853.3 ( 8.58x)
hevc_idct_32x32_8_rvv_i64: 20247.3 ( 5.92x)
(before)
Changes in v5:
Le 29 mai 2025 19:02:24 GMT+03:00, Michael Niedermayer
a écrit :
>can gcc or clang not build code like our runtime cpudetect ?
You can, on some versions and some architectures, select the target CPU per
function, but you can't select multiple targets, nor have the compiler
automatically sele
On Fri, 30 May 2025, Dmitriy Kovalenko wrote:
I've found quite a few ways to optimize existing ffmpeg's rgb to yuv
subsampled conversion. In this patch stack I'll try to
improve the perofrmance.
This particular set of changes is a small improvement to all the
existing functions and macro. The b
All the comments were addressed, except the prefetch one in the patch version 2
I sent earlier today. And how did you test the prefetch, because I literally
run a native benchmarking on the device right now and I see that with the patch
applied I am getting 5% of performance improvement. Maybe t
This patch integrates so called double bufferring when we are loading
2 batch of elements at a time and then processing them in parallel. On the
moden arm processors especially Apple Silicon it gives a visible
benefit, for subsampled pixel processing it is especially nice because
it allows to read
I've found quite a few ways to optimize existing ffmpeg's rgb to yuv
subsampled conversion. In this patch stack I'll try to
improve the perofrmance.
This particular set of changes is a small improvement to all the
existing functions and macro. The biggest performance gain is
coming from post loadi
On Fri, 30 May 2025, Dmitriy Kovalenko wrote:
All the comments were addressed
If you fail to read the inline responses, we can end this conversation
right here.
And how did you test the prefetch, because I literally run a native
benchmarking on the device right now and I see that with the
On Thu, 29 May 2025, Dmitriy Kovalenko wrote:
I appreciate the review for both the commits. I did fix all the
unrelated changes and iterated in the new version, would appreciate the
rearview.
Don't top post.
There are still at least 5 of my comments unaddressed. If you are not
going to addr
On Thu, 29 May 2025, Martin Storsjö wrote:
On Tue, 27 May 2025, Dmitriy Kovalenko wrote:
This patches integrates so called double bufferring when we are loading
2 batch elements at a time and then processing them in parallel. On the
moden arm processors especially Apple Silicon it gives a visi
On Thu, 29 May 2025, Martin Storsjö wrote:
In this case, they also add a direct dependence on
the updated pointer register from the directly preceding instruction, which
_is_ harmful on in-order cores, unless it entirely ignores the instruction.)
I did benchmark this, and indeed it causes a l
76 matches
Mail list logo
