diff mbox series

[FFmpeg-devel,v3] Add avpriv_slicethread_create2() and avpriv_slicethread_execute2(), make execute() and execute2() return FFMIN() of thread return codes

Message ID 99afc63012bcb654bba64f2cadc2d1404316c027.camel@haerdin.se
State New
Headers show
Series [FFmpeg-devel,v3] Add avpriv_slicethread_create2() and avpriv_slicethread_execute2(), make execute() and execute2() return FFMIN() of thread return codes | expand

Checks

Context Check Description
yinshiyou/configure_loongarch64 warning Failed to apply patch
andriy/commit_msg_x86 warning The first line of the commit message must start with a context terminated by a colon and a space, for example "lavu/opt: " or "doc: ".
andriy/make_x86 success Make finished
andriy/make_fate_x86 success Make fate finished

Commit Message

Tomas Härdin Sept. 26, 2022, 1:46 p.m. UTC
Fixed the ABI break in the previous version of this patch, updated all
internal uses of related APIs.

Tested with both ./configure and ./configure --disable-pthreads

Also using a new email for mailing list purposes.

/Tomas

Comments

Andreas Rheinhardt Sept. 26, 2022, 1:50 p.m. UTC | #1
Tomas Härdin:
> diff --git a/doc/APIchanges b/doc/APIchanges
> index b0a41c9e37..240e549a2f 100644
> --- a/doc/APIchanges
> +++ b/doc/APIchanges
> @@ -14,6 +14,10 @@ libavutil:     2021-04-27
>  
>  API changes, most recent first:
>  
> +2022-09-26 - xxxxxxxxxx - lavu 57.38.100 - slicethread.h
> +  Deprecate avpriv_slicethread_create() and avpriv_slicethread_execute().
> +  Add avpriv_slicethread_create2() and avpriv_slicethread_execute2().
> +

This is not public API and therefore no entry in APIchanges is necessary
or warranted.

>  2022-09-26 - xxxxxxxxxx - lavc 59.48.100 - avcodec.h
>    Deprecate avcodec_enum_to_chroma_pos() and avcodec_chroma_pos_to_enum().
>    Use av_chroma_location_enum_to_pos() or av_chroma_location_pos_to_enum()
James Almer Sept. 26, 2022, 1:54 p.m. UTC | #2
On 9/26/2022 10:50 AM, Andreas Rheinhardt wrote:
> Tomas Härdin:
>> diff --git a/doc/APIchanges b/doc/APIchanges
>> index b0a41c9e37..240e549a2f 100644
>> --- a/doc/APIchanges
>> +++ b/doc/APIchanges
>> @@ -14,6 +14,10 @@ libavutil:     2021-04-27
>>   
>>   API changes, most recent first:
>>   
>> +2022-09-26 - xxxxxxxxxx - lavu 57.38.100 - slicethread.h
>> +  Deprecate avpriv_slicethread_create() and avpriv_slicethread_execute().
>> +  Add avpriv_slicethread_create2() and avpriv_slicethread_execute2().
>> +
> 
> This is not public API and therefore no entry in APIchanges is necessary
> or warranted.

And for the same reason there's no need to use the deprecated attribute 
on them.

Also, maybe wrap all the old stuff in LIBAVUTIL_VERSION_MAJOR < 58 
checks so they are disabled gone as soon as we bump major, or at least 
make some noise so we don't forget about it.

> 
>>   2022-09-26 - xxxxxxxxxx - lavc 59.48.100 - avcodec.h
>>     Deprecate avcodec_enum_to_chroma_pos() and avcodec_chroma_pos_to_enum().
>>     Use av_chroma_location_enum_to_pos() or av_chroma_location_pos_to_enum()
> 
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel@ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
> 
> To unsubscribe, visit link above, or email
> ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
Andreas Rheinhardt Sept. 26, 2022, 1:58 p.m. UTC | #3
Tomas Härdin:
> diff --git a/libswscale/swscale_internal.h b/libswscale/swscale_internal.h
> index abeebbb002..50e73c86fa 100644
> --- a/libswscale/swscale_internal.h
> +++ b/libswscale/swscale_internal.h
> @@ -306,7 +306,7 @@ typedef struct SwsContext {
>  
>      AVSliceThread      *slicethread;
>      struct SwsContext **slice_ctx;
> -    int                *slice_err;
> +    attribute_deprecated int *slice_err; ///< @deprecated Not used any more. Removing it would require a bunch of asm to be rewritten.
>      int              nb_slice_ctx;
>  
>      // values passed to current sws_receive_slice() call

What asm would need to be rewritten? (I presume that no asm code
accesses slice_err, because doing so would be insane and because you
already stop allocating slice_err in this patch. Is it because of
hardcoded offsets of other fields? Which fields? Why is there no
corresponding AV_CHECK_OFFSET for them?)

Anyway, we don't deprecate internal fields; if we have to keep them
around for ABI compatibility, we wrap them in #if LIBFOO_VERSION_MAJOR <
BAR.

- Andreas
Andreas Rheinhardt Sept. 26, 2022, 2:01 p.m. UTC | #4
Andreas Rheinhardt:
> Tomas Härdin:
>> diff --git a/libswscale/swscale_internal.h b/libswscale/swscale_internal.h
>> index abeebbb002..50e73c86fa 100644
>> --- a/libswscale/swscale_internal.h
>> +++ b/libswscale/swscale_internal.h
>> @@ -306,7 +306,7 @@ typedef struct SwsContext {
>>  
>>      AVSliceThread      *slicethread;
>>      struct SwsContext **slice_ctx;
>> -    int                *slice_err;
>> +    attribute_deprecated int *slice_err; ///< @deprecated Not used any more. Removing it would require a bunch of asm to be rewritten.
>>      int              nb_slice_ctx;
>>  
>>      // values passed to current sws_receive_slice() call
> 
> What asm would need to be rewritten? (I presume that no asm code
> accesses slice_err, because doing so would be insane and because you
> already stop allocating slice_err in this patch. Is it because of
> hardcoded offsets of other fields? Which fields? Why is there no
> corresponding AV_CHECK_OFFSET for them?)
> 
> Anyway, we don't deprecate internal fields; if we have to keep them
> around for ABI compatibility, we wrap them in #if LIBFOO_VERSION_MAJOR <
> BAR.
> 

Ok, just moving this led me to the issue: An assert in line 751 of
libswscale/x86/swscale.c.

This is actually very frightening. A dev who does not use x86 could
break x86 any time.

- Andreas
Tomas Härdin Sept. 26, 2022, 2:02 p.m. UTC | #5
mån 2022-09-26 klockan 15:58 +0200 skrev Andreas Rheinhardt:
> Tomas Härdin:
> > diff --git a/libswscale/swscale_internal.h
> > b/libswscale/swscale_internal.h
> > index abeebbb002..50e73c86fa 100644
> > --- a/libswscale/swscale_internal.h
> > +++ b/libswscale/swscale_internal.h
> > @@ -306,7 +306,7 @@ typedef struct SwsContext {
> >  
> >      AVSliceThread      *slicethread;
> >      struct SwsContext **slice_ctx;
> > -    int                *slice_err;
> > +    attribute_deprecated int *slice_err; ///< @deprecated Not used
> > any more. Removing it would require a bunch of asm to be rewritten.
> >      int              nb_slice_ctx;
> >  
> >      // values passed to current sws_receive_slice() call
> 
> What asm would need to be rewritten?

swscale.c:
        /* yuv2gbrp uses the SwsContext for yuv coefficients
           if struct offsets change the asm needs to be updated too */
        av_assert0(offsetof(SwsContext, yuv2rgb_y_offset) == 40292);


> Anyway, we don't deprecate internal fields; if we have to keep them
> around for ABI compatibility, we wrap them in #if
> LIBFOO_VERSION_MAJOR <
> BAR.

It's not actually an API/ABI issue but an asm issue

/Tomas
Tomas Härdin Sept. 26, 2022, 2:14 p.m. UTC | #6
mån 2022-09-26 klockan 10:54 -0300 skrev James Almer:
> On 9/26/2022 10:50 AM, Andreas Rheinhardt wrote:
> > Tomas Härdin:
> > > diff --git a/doc/APIchanges b/doc/APIchanges
> > > index b0a41c9e37..240e549a2f 100644
> > > --- a/doc/APIchanges
> > > +++ b/doc/APIchanges
> > > @@ -14,6 +14,10 @@ libavutil:     2021-04-27
> > >   
> > >   API changes, most recent first:
> > >   
> > > +2022-09-26 - xxxxxxxxxx - lavu 57.38.100 - slicethread.h
> > > +  Deprecate avpriv_slicethread_create() and
> > > avpriv_slicethread_execute().
> > > +  Add avpriv_slicethread_create2() and
> > > avpriv_slicethread_execute2().
> > > +
> > 
> > This is not public API and therefore no entry in APIchanges is
> > necessary
> > or warranted.
> 
> And for the same reason there's no need to use the deprecated
> attribute 
> on them.
> 
> Also, maybe wrap all the old stuff in LIBAVUTIL_VERSION_MAJOR < 58 
> checks so they are disabled gone as soon as we bump major, or at
> least 
> make some noise so we don't forget about it.

Updated patch attached

/Tomas
Andreas Rheinhardt Sept. 26, 2022, 2:28 p.m. UTC | #7
Tomas Härdin:
> mån 2022-09-26 klockan 15:58 +0200 skrev Andreas Rheinhardt:
>> Tomas Härdin:
>>> diff --git a/libswscale/swscale_internal.h
>>> b/libswscale/swscale_internal.h
>>> index abeebbb002..50e73c86fa 100644
>>> --- a/libswscale/swscale_internal.h
>>> +++ b/libswscale/swscale_internal.h
>>> @@ -306,7 +306,7 @@ typedef struct SwsContext {
>>>  
>>>      AVSliceThread      *slicethread;
>>>      struct SwsContext **slice_ctx;
>>> -    int                *slice_err;
>>> +    attribute_deprecated int *slice_err; ///< @deprecated Not used
>>> any more. Removing it would require a bunch of asm to be rewritten.
>>>      int              nb_slice_ctx;
>>>  
>>>      // values passed to current sws_receive_slice() call
>>
>> What asm would need to be rewritten?
> 
> swscale.c:
>         /* yuv2gbrp uses the SwsContext for yuv coefficients
>            if struct offsets change the asm needs to be updated too */
>         av_assert0(offsetof(SwsContext, yuv2rgb_y_offset) == 40292);
> 
> 
>> Anyway, we don't deprecate internal fields; if we have to keep them
>> around for ABI compatibility, we wrap them in #if
>> LIBFOO_VERSION_MAJOR <
>> BAR.
> 
> It's not actually an API/ABI issue but an asm issue
> 

It seems to me that this is the only such issue that affects this patch.
If you modified line 583 of libswscale/x86/output.asm as well as the
assert, you could remove slice_err. (But let James confirm this.)

- Andreas
Tomas Härdin Sept. 26, 2022, 2:44 p.m. UTC | #8
mån 2022-09-26 klockan 16:28 +0200 skrev Andreas Rheinhardt:
> Tomas Härdin:
> > mån 2022-09-26 klockan 15:58 +0200 skrev Andreas Rheinhardt:
> > > Tomas Härdin:
> > > > diff --git a/libswscale/swscale_internal.h
> > > > b/libswscale/swscale_internal.h
> > > > index abeebbb002..50e73c86fa 100644
> > > > --- a/libswscale/swscale_internal.h
> > > > +++ b/libswscale/swscale_internal.h
> > > > @@ -306,7 +306,7 @@ typedef struct SwsContext {
> > > >  
> > > >      AVSliceThread      *slicethread;
> > > >      struct SwsContext **slice_ctx;
> > > > -    int                *slice_err;
> > > > +    attribute_deprecated int *slice_err; ///< @deprecated Not
> > > > used
> > > > any more. Removing it would require a bunch of asm to be
> > > > rewritten.
> > > >      int              nb_slice_ctx;
> > > >  
> > > >      // values passed to current sws_receive_slice() call
> > > 
> > > What asm would need to be rewritten?
> > 
> > swscale.c:
> >         /* yuv2gbrp uses the SwsContext for yuv coefficients
> >            if struct offsets change the asm needs to be updated too
> > */
> >         av_assert0(offsetof(SwsContext, yuv2rgb_y_offset) ==
> > 40292);
> > 
> > 
> > > Anyway, we don't deprecate internal fields; if we have to keep
> > > them
> > > around for ABI compatibility, we wrap them in #if
> > > LIBFOO_VERSION_MAJOR <
> > > BAR.
> > 
> > It's not actually an API/ABI issue but an asm issue
> > 
> 
> It seems to me that this is the only such issue that affects this
> patch.
> If you modified line 583 of libswscale/x86/output.asm as well as the
> assert, you could remove slice_err. (But let James confirm this.)

Here's a yet further updated patch with this change implemented. Passes
FATE.

/Tomas
diff mbox series

Patch

From 265f1095fb85b1eba18ed1b89fe71531ee09fe9b Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Tomas=20H=C3=A4rdin?= <git@haerdin.se>
Date: Thu, 16 Jun 2022 12:16:44 +0200
Subject: [PATCH] Add avpriv_slicethread_create2() and
 avpriv_slicethread_execute2(), make execute() and execute2() return FFMIN()
 of thread return codes

Updates all internal uses of related APIs.
At the moment only fic.c actually checks return code of execute() hence the change to its FATE reference.
---
 doc/APIchanges                |  4 ++
 libavcodec/avcodec.c          | 10 +++--
 libavcodec/pthread_slice.c    | 14 +++----
 libavfilter/pthread.c         |  8 ++--
 libavutil/slicethread.c       | 70 ++++++++++++++++++++++++++++++++---
 libavutil/slicethread.h       | 27 ++++++++++++++
 libavutil/version.h           |  2 +-
 libswscale/swscale.c          | 20 ++--------
 libswscale/swscale_internal.h |  6 +--
 libswscale/utils.c            |  8 ++--
 tests/fate/screen.mak         |  1 +
 tests/ref/fate/fic-avi        | 30 +++++++--------
 12 files changed, 137 insertions(+), 63 deletions(-)

diff --git a/doc/APIchanges b/doc/APIchanges
index b0a41c9e37..240e549a2f 100644
--- a/doc/APIchanges
+++ b/doc/APIchanges
@@ -14,6 +14,10 @@  libavutil:     2021-04-27
 
 API changes, most recent first:
 
+2022-09-26 - xxxxxxxxxx - lavu 57.38.100 - slicethread.h
+  Deprecate avpriv_slicethread_create() and avpriv_slicethread_execute().
+  Add avpriv_slicethread_create2() and avpriv_slicethread_execute2().
+
 2022-09-26 - xxxxxxxxxx - lavc 59.48.100 - avcodec.h
   Deprecate avcodec_enum_to_chroma_pos() and avcodec_chroma_pos_to_enum().
   Use av_chroma_location_enum_to_pos() or av_chroma_location_pos_to_enum()
diff --git a/libavcodec/avcodec.c b/libavcodec/avcodec.c
index a85d3c2309..ee75b3abe7 100644
--- a/libavcodec/avcodec.c
+++ b/libavcodec/avcodec.c
@@ -44,28 +44,30 @@ 
 
 int avcodec_default_execute(AVCodecContext *c, int (*func)(AVCodecContext *c2, void *arg2), void *arg, int *ret, int count, int size)
 {
-    int i;
+    int i, rr = 0;
 
     for (i = 0; i < count; i++) {
         int r = func(c, (char *)arg + i * size);
+        rr = FFMIN(rr, r);
         if (ret)
             ret[i] = r;
     }
     emms_c();
-    return 0;
+    return rr;
 }
 
 int avcodec_default_execute2(AVCodecContext *c, int (*func)(AVCodecContext *c2, void *arg2, int jobnr, int threadnr), void *arg, int *ret, int count)
 {
-    int i;
+    int i, rr = 0;
 
     for (i = 0; i < count; i++) {
         int r = func(c, arg, i, 0);
+        rr = FFMIN(rr, r);
         if (ret)
             ret[i] = r;
     }
     emms_c();
-    return 0;
+    return rr;
 }
 
 static AVMutex codec_mutex = AV_MUTEX_INITIALIZER;
diff --git a/libavcodec/pthread_slice.c b/libavcodec/pthread_slice.c
index a4d31c6f4d..795d0c9c5a 100644
--- a/libavcodec/pthread_slice.c
+++ b/libavcodec/pthread_slice.c
@@ -61,13 +61,13 @@  typedef struct SliceThreadContext {
     Progress *progress;
 } SliceThreadContext;
 
-static void main_function(void *priv) {
+static int main_function(void *priv) {
     AVCodecContext *avctx = priv;
     SliceThreadContext *c = avctx->internal->thread_ctx;
-    c->mainfunc(avctx);
+    return c->mainfunc(avctx);
 }
 
-static void worker_func(void *priv, int jobnr, int threadnr, int nb_jobs, int nb_threads)
+static int worker_func(void *priv, int jobnr, int threadnr, int nb_jobs, int nb_threads)
 {
     AVCodecContext *avctx = priv;
     SliceThreadContext *c = avctx->internal->thread_ctx;
@@ -77,6 +77,7 @@  static void worker_func(void *priv, int jobnr, int threadnr, int nb_jobs, int nb
                   : c->func2(avctx, c->args, jobnr, threadnr);
     if (c->rets)
         c->rets[jobnr] = ret;
+    return ret;
 }
 
 void ff_slice_thread_free(AVCodecContext *avctx)
@@ -112,8 +113,7 @@  static int thread_execute(AVCodecContext *avctx, action_func* func, void *arg, i
     c->func = func;
     c->rets = ret;
 
-    avpriv_slicethread_execute(c->thread, job_count, !!c->mainfunc  );
-    return 0;
+    return avpriv_slicethread_execute2(c->thread, job_count, !!c->mainfunc);
 }
 
 static int thread_execute2(AVCodecContext *avctx, action_func2* func2, void *arg, int *ret, int job_count)
@@ -135,7 +135,7 @@  int ff_slice_thread_init(AVCodecContext *avctx)
 {
     SliceThreadContext *c;
     int thread_count = avctx->thread_count;
-    void (*mainfunc)(void *);
+    int (*mainfunc)(void *);
 
     // We cannot do this in the encoder init as the threads are created before
     if (av_codec_is_encoder(avctx->codec) &&
@@ -161,7 +161,7 @@  int ff_slice_thread_init(AVCodecContext *avctx)
 
     avctx->internal->thread_ctx = c = av_mallocz(sizeof(*c));
     mainfunc = ffcodec(avctx->codec)->caps_internal & FF_CODEC_CAP_SLICE_THREAD_HAS_MF ? &main_function : NULL;
-    if (!c || (thread_count = avpriv_slicethread_create(&c->thread, avctx, worker_func, mainfunc, thread_count)) <= 1) {
+    if (!c || (thread_count = avpriv_slicethread_create2(&c->thread, avctx, worker_func, mainfunc, thread_count)) <= 1) {
         if (c)
             avpriv_slicethread_free(&c->thread);
         av_freep(&avctx->internal->thread_ctx);
diff --git a/libavfilter/pthread.c b/libavfilter/pthread.c
index 1a063d3cc0..855c842df9 100644
--- a/libavfilter/pthread.c
+++ b/libavfilter/pthread.c
@@ -43,12 +43,13 @@  typedef struct ThreadContext {
     int   *rets;
 } ThreadContext;
 
-static void worker_func(void *priv, int jobnr, int threadnr, int nb_jobs, int nb_threads)
+static int worker_func(void *priv, int jobnr, int threadnr, int nb_jobs, int nb_threads)
 {
     ThreadContext *c = priv;
     int ret = c->func(c->ctx, c->arg, jobnr, nb_jobs);
     if (c->rets)
         c->rets[jobnr] = ret;
+    return ret;
 }
 
 static void slice_thread_uninit(ThreadContext *c)
@@ -68,13 +69,12 @@  static int thread_execute(AVFilterContext *ctx, avfilter_action_func *func,
     c->func        = func;
     c->rets        = ret;
 
-    avpriv_slicethread_execute(c->thread, nb_jobs, 0);
-    return 0;
+    return avpriv_slicethread_execute2(c->thread, nb_jobs, 0);
 }
 
 static int thread_init_internal(ThreadContext *c, int nb_threads)
 {
-    nb_threads = avpriv_slicethread_create(&c->thread, c, worker_func, NULL, nb_threads);
+    nb_threads = avpriv_slicethread_create2(&c->thread, c, worker_func, NULL, nb_threads);
     if (nb_threads <= 1)
         avpriv_slicethread_free(&c->thread);
     return FFMAX(nb_threads, 1);
diff --git a/libavutil/slicethread.c b/libavutil/slicethread.c
index 115b099736..38c4ebdc4c 100644
--- a/libavutil/slicethread.c
+++ b/libavutil/slicethread.c
@@ -34,6 +34,7 @@  typedef struct WorkerContext {
     pthread_cond_t  cond;
     pthread_t       thread;
     int             done;
+    int             ret;
 } WorkerContext;
 
 struct AVSliceThread {
@@ -50,11 +51,15 @@  struct AVSliceThread {
     int             finished;
 
     void            *priv;
+    // either worker_func and main_func are non-NULL or worker_func2 and main_func2 are
+    // all four are never non-NULL at the same time
     void            (*worker_func)(void *priv, int jobnr, int threadnr, int nb_jobs, int nb_threads);
     void            (*main_func)(void *priv);
+    int             (*worker_func2)(void *priv, int jobnr, int threadnr, int nb_jobs, int nb_threads);
+    int             (*main_func2)(void *priv);
 };
 
-static int run_jobs(AVSliceThread *ctx)
+static int run_jobs(AVSliceThread *ctx, int *ret_out)
 {
     unsigned nb_jobs    = ctx->nb_jobs;
     unsigned nb_active_threads = ctx->nb_active_threads;
@@ -62,7 +67,12 @@  static int run_jobs(AVSliceThread *ctx)
     unsigned current_job  = first_job;
 
     do {
-        ctx->worker_func(ctx->priv, current_job, first_job, nb_jobs, nb_active_threads);
+        if (ctx->worker_func2) {
+            int ret = ctx->worker_func2(ctx->priv, current_job, first_job, nb_jobs, nb_active_threads);
+            *ret_out = FFMIN(*ret_out, ret);
+        } else {
+            ctx->worker_func(ctx->priv, current_job, first_job, nb_jobs, nb_active_threads);
+        }
     } while ((current_job = atomic_fetch_add_explicit(&ctx->current_job, 1, memory_order_acq_rel)) < nb_jobs);
 
     return current_job == nb_jobs + nb_active_threads - 1;
@@ -86,7 +96,7 @@  static void *attribute_align_arg thread_worker(void *v)
             return NULL;
         }
 
-        if (run_jobs(ctx)) {
+        if (run_jobs(ctx, &w->ret)) {
             pthread_mutex_lock(&ctx->done_mutex);
             ctx->done = 1;
             pthread_cond_signal(&ctx->done_cond);
@@ -95,9 +105,11 @@  static void *attribute_align_arg thread_worker(void *v)
     }
 }
 
-int avpriv_slicethread_create(AVSliceThread **pctx, void *priv,
+static int slicethread_create(AVSliceThread **pctx, void *priv,
                               void (*worker_func)(void *priv, int jobnr, int threadnr, int nb_jobs, int nb_threads),
                               void (*main_func)(void *priv),
+                              int (*worker_func2)(void *priv, int jobnr, int threadnr, int nb_jobs, int nb_threads),
+                              int (*main_func2)(void *priv),
                               int nb_threads)
 {
     AVSliceThread *ctx;
@@ -128,6 +140,8 @@  int avpriv_slicethread_create(AVSliceThread **pctx, void *priv,
     ctx->priv        = priv;
     ctx->worker_func = worker_func;
     ctx->main_func   = main_func;
+    ctx->worker_func2= worker_func2;
+    ctx->main_func2  = main_func2;
     ctx->nb_threads  = nb_threads;
     ctx->nb_active_threads = 0;
     ctx->nb_jobs     = 0;
@@ -165,9 +179,30 @@  int avpriv_slicethread_create(AVSliceThread **pctx, void *priv,
     return nb_threads;
 }
 
+int avpriv_slicethread_create(AVSliceThread **pctx, void *priv,
+                              void (*worker_func)(void *priv, int jobnr, int threadnr, int nb_jobs, int nb_threads),
+                              void (*main_func)(void *priv),
+                              int nb_threads)
+{
+    return slicethread_create(pctx, priv, worker_func, main_func, NULL, NULL, nb_threads);
+}
+
+int avpriv_slicethread_create2(AVSliceThread **pctx, void *priv,
+                               int (*worker_func)(void *priv, int jobnr, int threadnr, int nb_jobs, int nb_threads),
+                               int (*main_func)(void *priv),
+                               int nb_threads)
+{
+    return slicethread_create(pctx, priv, NULL, NULL, worker_func, main_func, nb_threads);
+}
+
 void avpriv_slicethread_execute(AVSliceThread *ctx, int nb_jobs, int execute_main)
 {
-    int nb_workers, i, is_last = 0;
+    (void)avpriv_slicethread_execute2(ctx, nb_jobs, execute_main);
+}
+
+int avpriv_slicethread_execute2(AVSliceThread *ctx, int nb_jobs, int execute_main)
+{
+    int nb_workers, i, is_last = 0, ret = 0;
 
     av_assert0(nb_jobs > 0);
     ctx->nb_jobs           = nb_jobs;
@@ -182,14 +217,17 @@  void avpriv_slicethread_execute(AVSliceThread *ctx, int nb_jobs, int execute_mai
         WorkerContext *w = &ctx->workers[i];
         pthread_mutex_lock(&w->mutex);
         w->done = 0;
+        w->ret = 0;
         pthread_cond_signal(&w->cond);
         pthread_mutex_unlock(&w->mutex);
     }
 
     if (ctx->main_func && execute_main)
         ctx->main_func(ctx->priv);
+    else if (ctx->main_func2 && execute_main)
+        ret = ctx->main_func2(ctx->priv);
     else
-        is_last = run_jobs(ctx);
+        is_last = run_jobs(ctx, &ret);
 
     if (!is_last) {
         pthread_mutex_lock(&ctx->done_mutex);
@@ -198,6 +236,11 @@  void avpriv_slicethread_execute(AVSliceThread *ctx, int nb_jobs, int execute_mai
         ctx->done = 0;
         pthread_mutex_unlock(&ctx->done_mutex);
     }
+
+    for (i = 0; i < nb_workers; i++)
+        ret = FFMIN(ret, ctx->workers[i].ret);
+
+    return ret;
 }
 
 void avpriv_slicethread_free(AVSliceThread **pctx)
@@ -246,11 +289,26 @@  int avpriv_slicethread_create(AVSliceThread **pctx, void *priv,
     return AVERROR(ENOSYS);
 }
 
+int avpriv_slicethread_create2(AVSliceThread **pctx, void *priv,
+                               int (*worker_func)(void *priv, int jobnr, int threadnr, int nb_jobs, int nb_threads),
+                               int (*main_func)(void *priv),
+                               int nb_threads)
+{
+    *pctx = NULL;
+    return AVERROR(ENOSYS);
+}
+
 void avpriv_slicethread_execute(AVSliceThread *ctx, int nb_jobs, int execute_main)
 {
     av_assert0(0);
 }
 
+int avpriv_slicethread_execute2(AVSliceThread *ctx, int nb_jobs, int execute_main)
+{
+    av_assert0(0);
+    return AVERROR(ENOSYS);
+}
+
 void avpriv_slicethread_free(AVSliceThread **pctx)
 {
     av_assert0(!pctx || !*pctx);
diff --git a/libavutil/slicethread.h b/libavutil/slicethread.h
index f6f6f302c4..4deb287674 100644
--- a/libavutil/slicethread.h
+++ b/libavutil/slicethread.h
@@ -29,20 +29,47 @@  typedef struct AVSliceThread AVSliceThread;
  * @param main_func special callback function, called from main thread, may be NULL
  * @param nb_threads number of threads, 0 for automatic, must be >= 0
  * @return return number of threads or negative AVERROR on failure
+ * @deprecated Use avpriv_slicethread_create2() instead.
  */
+attribute_deprecated
 int avpriv_slicethread_create(AVSliceThread **pctx, void *priv,
                               void (*worker_func)(void *priv, int jobnr, int threadnr, int nb_jobs, int nb_threads),
                               void (*main_func)(void *priv),
                               int nb_threads);
 
+/**
+ * Like avpriv_slicethread_create() except worker_func and main_func return int.
+ * @param pctx slice threading context returned here
+ * @param priv private pointer to be passed to callback function
+ * @param worker_func callback function to be executed
+ * @param main_func special callback function, called from main thread, may be NULL
+ * @param nb_threads number of threads, 0 for automatic, must be >= 0
+ * @return return number of threads or negative AVERROR on failure
+ */
+int avpriv_slicethread_create2(AVSliceThread **pctx, void *priv,
+                               int (*worker_func)(void *priv, int jobnr, int threadnr, int nb_jobs, int nb_threads),
+                               int (*main_func)(void *priv),
+                               int nb_threads);
+
 /**
  * Execute slice threading.
  * @param ctx slice threading context
  * @param nb_jobs number of jobs, must be > 0
  * @param execute_main also execute main_func
+ * @deprecated Use avpriv_slicethread_execute2() instead.
  */
+attribute_deprecated
 void avpriv_slicethread_execute(AVSliceThread *ctx, int nb_jobs, int execute_main);
 
+/**
+ * Like avpriv_slicethread_execute() except returns FFMIN(0, main_func(), worker_func()).
+ * @param ctx slice threading context
+ * @param nb_jobs number of jobs, must be > 0
+ * @param execute_main also execute main_func
+ * @return AVERROR on error.
+ */
+int avpriv_slicethread_execute2(AVSliceThread *ctx, int nb_jobs, int execute_main);
+
 /**
  * Destroy slice threading context.
  * @param pctx pointer to context
diff --git a/libavutil/version.h b/libavutil/version.h
index 9c44cef6aa..5aca550f45 100644
--- a/libavutil/version.h
+++ b/libavutil/version.h
@@ -79,7 +79,7 @@ 
  */
 
 #define LIBAVUTIL_VERSION_MAJOR  57
-#define LIBAVUTIL_VERSION_MINOR  37
+#define LIBAVUTIL_VERSION_MINOR  38
 #define LIBAVUTIL_VERSION_MICRO 100
 
 #define LIBAVUTIL_VERSION_INT   AV_VERSION_INT(LIBAVUTIL_VERSION_MAJOR, \
diff --git a/libswscale/swscale.c b/libswscale/swscale.c
index 367d045a02..500b74c099 100644
--- a/libswscale/swscale.c
+++ b/libswscale/swscale.c
@@ -1150,23 +1150,11 @@  int sws_receive_slice(struct SwsContext *c, unsigned int slice_start,
 
     if (c->slicethread) {
         int nb_jobs = c->slice_ctx[0]->dither == SWS_DITHER_ED ? 1 : c->nb_slice_ctx;
-        int ret = 0;
 
         c->dst_slice_start  = slice_start;
         c->dst_slice_height = slice_height;
 
-        avpriv_slicethread_execute(c->slicethread, nb_jobs, 0);
-
-        for (int i = 0; i < c->nb_slice_ctx; i++) {
-            if (c->slice_err[i] < 0) {
-                ret = c->slice_err[i];
-                break;
-            }
-        }
-
-        memset(c->slice_err, 0, c->nb_slice_ctx * sizeof(*c->slice_err));
-
-        return ret;
+        return avpriv_slicethread_execute2(c->slicethread, nb_jobs, 0);
     }
 
     for (int i = 0; i < FF_ARRAY_ELEMS(dst); i++) {
@@ -1213,8 +1201,8 @@  int attribute_align_arg sws_scale(struct SwsContext *c,
                           dst, dstStride, 0, c->dstH);
 }
 
-void ff_sws_slice_worker(void *priv, int jobnr, int threadnr,
-                         int nb_jobs, int nb_threads)
+int ff_sws_slice_worker(void *priv, int jobnr, int threadnr,
+                        int nb_jobs, int nb_threads)
 {
     SwsContext *parent = priv;
     SwsContext      *c = parent->slice_ctx[threadnr];
@@ -1242,5 +1230,5 @@  void ff_sws_slice_worker(void *priv, int jobnr, int threadnr,
                              parent->dst_slice_start + slice_start, slice_end - slice_start);
     }
 
-    parent->slice_err[threadnr] = err;
+    return err;
 }
diff --git a/libswscale/swscale_internal.h b/libswscale/swscale_internal.h
index abeebbb002..50e73c86fa 100644
--- a/libswscale/swscale_internal.h
+++ b/libswscale/swscale_internal.h
@@ -306,7 +306,7 @@  typedef struct SwsContext {
 
     AVSliceThread      *slicethread;
     struct SwsContext **slice_ctx;
-    int                *slice_err;
+    attribute_deprecated int *slice_err; ///< @deprecated Not used any more. Removing it would require a bunch of asm to be rewritten.
     int              nb_slice_ctx;
 
     // values passed to current sws_receive_slice() call
@@ -1152,8 +1152,8 @@  void ff_init_vscale_pfn(SwsContext *c, yuv2planar1_fn yuv2plane1, yuv2planarX_fn
     yuv2interleavedX_fn yuv2nv12cX, yuv2packed1_fn yuv2packed1, yuv2packed2_fn yuv2packed2,
     yuv2packedX_fn yuv2packedX, yuv2anyX_fn yuv2anyX, int use_mmx);
 
-void ff_sws_slice_worker(void *priv, int jobnr, int threadnr,
-                         int nb_jobs, int nb_threads);
+int ff_sws_slice_worker(void *priv, int jobnr, int threadnr,
+                        int nb_jobs, int nb_threads);
 
 //number of extra lines to process
 #define MAX_LINES_AHEAD 4
diff --git a/libswscale/utils.c b/libswscale/utils.c
index 45baa22b23..8bfd3a64ac 100644
--- a/libswscale/utils.c
+++ b/libswscale/utils.c
@@ -1273,8 +1273,8 @@  static int context_init_threaded(SwsContext *c,
 {
     int ret;
 
-    ret = avpriv_slicethread_create(&c->slicethread, (void*)c,
-                                    ff_sws_slice_worker, NULL, c->nb_threads);
+    ret = avpriv_slicethread_create2(&c->slicethread, (void*)c,
+                                     ff_sws_slice_worker, NULL, c->nb_threads);
     if (ret == AVERROR(ENOSYS)) {
         c->nb_threads = 1;
         return 0;
@@ -1284,8 +1284,7 @@  static int context_init_threaded(SwsContext *c,
     c->nb_threads = ret;
 
     c->slice_ctx = av_calloc(c->nb_threads, sizeof(*c->slice_ctx));
-    c->slice_err = av_calloc(c->nb_threads, sizeof(*c->slice_err));
-    if (!c->slice_ctx || !c->slice_err)
+    if (!c->slice_ctx)
         return AVERROR(ENOMEM);
 
     for (int i = 0; i < c->nb_threads; i++) {
@@ -2421,7 +2420,6 @@  void sws_freeContext(SwsContext *c)
     for (i = 0; i < c->nb_slice_ctx; i++)
         sws_freeContext(c->slice_ctx[i]);
     av_freep(&c->slice_ctx);
-    av_freep(&c->slice_err);
 
     avpriv_slicethread_free(&c->slicethread);
 
diff --git a/tests/fate/screen.mak b/tests/fate/screen.mak
index bd6d228544..b572f67b55 100644
--- a/tests/fate/screen.mak
+++ b/tests/fate/screen.mak
@@ -5,6 +5,7 @@  fate-cscd: CMD = framecrc -i $(TARGET_SAMPLES)/CSCD/sample_video.avi -an -pix_fm
 FATE_SCREEN-$(call FRAMECRC, AVI, DXTORY) += fate-dxtory
 fate-dxtory: CMD = framecrc -i $(TARGET_SAMPLES)/dxtory/dxtory_mic.avi -an
 
+# fic sample is broken. some packets are truncated and one packet has length zero
 FATE_SCREEN-$(call FRAMECRC, AVI, FIC) += fate-fic-avi
 fate-fic-avi: CMD = framecrc -i $(TARGET_SAMPLES)/fic/fic-partial-2MB.avi -an
 
diff --git a/tests/ref/fate/fic-avi b/tests/ref/fate/fic-avi
index df55789d54..4546f230b1 100644
--- a/tests/ref/fate/fic-avi
+++ b/tests/ref/fate/fic-avi
@@ -76,19 +76,18 @@ 
 0,         70,         70,        1,  1566720, 0x40f7d39a
 0,         71,         71,        1,  1566720, 0x40f7d39a
 0,         72,         72,        1,  1566720, 0x40f7d39a
-0,         73,         73,        1,  1566720, 0xa7d6e25f
-0,         74,         74,        1,  1566720, 0xa7d6e25f
-0,         75,         75,        1,  1566720, 0xa7d6e25f
-0,         76,         76,        1,  1566720, 0xa7d6e25f
-0,         77,         77,        1,  1566720, 0xa7d6e25f
-0,         78,         78,        1,  1566720, 0xa7d6e25f
-0,         79,         79,        1,  1566720, 0xa7d6e25f
-0,         80,         80,        1,  1566720, 0xa7d6e25f
-0,         81,         81,        1,  1566720, 0xa7d6e25f
-0,         82,         82,        1,  1566720, 0xa7d6e25f
-0,         83,         83,        1,  1566720, 0xa7d6e25f
-0,         84,         84,        1,  1566720, 0xa7d6e25f
-0,         85,         85,        1,  1566720, 0xa7d6e25f
+0,         74,         74,        1,  1566720, 0x40f7d39a
+0,         75,         75,        1,  1566720, 0x40f7d39a
+0,         76,         76,        1,  1566720, 0x40f7d39a
+0,         77,         77,        1,  1566720, 0x40f7d39a
+0,         78,         78,        1,  1566720, 0x40f7d39a
+0,         79,         79,        1,  1566720, 0x40f7d39a
+0,         80,         80,        1,  1566720, 0x40f7d39a
+0,         81,         81,        1,  1566720, 0x40f7d39a
+0,         82,         82,        1,  1566720, 0x40f7d39a
+0,         83,         83,        1,  1566720, 0x40f7d39a
+0,         84,         84,        1,  1566720, 0x40f7d39a
+0,         85,         85,        1,  1566720, 0x40f7d39a
 0,         86,         86,        1,  1566720, 0xa7d6e25f
 0,         87,         87,        1,  1566720, 0xa7d6e25f
 0,         88,         88,        1,  1566720, 0xa7d6e25f
@@ -104,7 +103,6 @@ 
 0,         98,         98,        1,  1566720, 0xa7d6e25f
 0,         99,         99,        1,  1566720, 0xa7d6e25f
 0,        100,        100,        1,  1566720, 0xeaf8d207
-0,        101,        101,        1,  1566720, 0x6724983e
 0,        102,        102,        1,  1566720, 0x0e95d209
 0,        103,        103,        1,  1566720, 0x0e95d209
 0,        104,        104,        1,  1566720, 0x0e95d209
@@ -121,6 +119,4 @@ 
 0,        115,        115,        1,  1566720, 0xfe83b964
 0,        116,        116,        1,  1566720, 0xfe83b964
 0,        117,        117,        1,  1566720, 0xfe83b964
-0,        118,        118,        1,  1566720, 0x25dc30a6
-0,        119,        119,        1,  1566720, 0x25dc30a6
-0,        120,        120,        1,  1566720, 0x25dc30a6
+0,        119,        119,        1,  1566720, 0xfe83b964
-- 
2.30.2