From 5221c25be67ee96a747c4e527436c0daa394a44e Mon Sep 17 00:00:00 2001 From: Vito Caputo Date: Fri, 28 Jul 2023 15:20:58 -0700 Subject: til: introduce til_module_render_limited() variant With the introduction of discoverable on-stream contexts, with the intention of modules finding them at runtime for nested rendering, relying exclusively on limiting n_cpus @ til_module_create_context() is no longer adequate. When the nested rendering makes use of a context created elsewhere, it can't make any assumptions about what its n_cpus is, nor should it be attempting to change that value for its use. So this variant just adds a max_cpus parameter for setting an upper bound to whatever is found in the context's n_cpus. The impetus for this is to fix the ref builtin, which you can currently trick into deadlock by performing a nested threaded render within a threaded render. What the ref module should be doing is propagating its own context's n_cpus down as the upper bound, via this new render variant. --- src/til.c | 30 +++++++++++++++++++++++------- 1 file changed, 23 insertions(+), 7 deletions(-) (limited to 'src/til.c') diff --git a/src/til.c b/src/til.c index d374759..b4a903c 100644 --- a/src/til.c +++ b/src/til.c @@ -386,7 +386,7 @@ char * til_get_module_names(unsigned flags_excluded, const char **exclusions) } -static void module_render_fragment(til_module_context_t *context, til_stream_t *stream, til_threads_t *threads, unsigned ticks, til_fb_fragment_t **fragment_ptr) +static void module_render_fragment(til_module_context_t *context, til_stream_t *stream, til_threads_t *threads, unsigned n_cpus, unsigned ticks, til_fb_fragment_t **fragment_ptr) { const til_module_t *module; int touched = 0; @@ -395,6 +395,7 @@ static void module_render_fragment(til_module_context_t *context, til_stream_t * assert(context->module); assert(threads); assert(fragment_ptr && *fragment_ptr); + assert(n_cpus); module = context->module; @@ -409,7 +410,7 @@ static void module_render_fragment(til_module_context_t *context, til_stream_t * assert(frame_plan.fragmenter); assert(module->render_fragment); - if (context->n_cpus > 1) { + if (n_cpus > 1) { til_threads_frame_submit(threads, fragment_ptr, &frame_plan, module->render_fragment, context, stream, ticks); til_threads_wait_idle(threads); } else { @@ -437,14 +438,11 @@ static void module_render_fragment(til_module_context_t *context, til_stream_t * } -/* This is a public interface to the threaded module rendering intended for use by - * modules that wish to get the output of other modules for their own use. - */ -void til_module_render(til_module_context_t *context, til_stream_t *stream, unsigned ticks, til_fb_fragment_t **fragment_ptr) +static void _til_module_render(til_module_context_t *context, til_stream_t *stream, unsigned n_cpus, unsigned ticks, til_fb_fragment_t **fragment_ptr) { unsigned start = til_ticks_now(); - module_render_fragment(context, stream, til_threads, ticks, fragment_ptr); + module_render_fragment(context, stream, til_threads, n_cpus, ticks, fragment_ptr); context->last_render_duration = til_ticks_now() - start; if (context->last_render_duration > context->max_render_duration) @@ -454,6 +452,24 @@ void til_module_render(til_module_context_t *context, til_stream_t *stream, unsi } +/* This is a public interface to the threaded module rendering intended for use by + * modules that wish to get the output of other modules for their own use. + */ +void til_module_render(til_module_context_t *context, til_stream_t *stream, unsigned ticks, til_fb_fragment_t **fragment_ptr) +{ + return _til_module_render(context, stream, context->n_cpus, ticks, fragment_ptr); +} + + +/* Identical to til_module_render() except with a parameterized upper bound for context->c_cpus, + * this is primarily intended for modules performing nested rendering + */ +void til_module_render_limited(til_module_context_t *context, til_stream_t *stream, unsigned ticks, unsigned max_cpus, til_fb_fragment_t **fragment_ptr) +{ + return _til_module_render(context, stream, MIN(context->n_cpus, max_cpus), ticks, fragment_ptr); +} + + /* if n_cpus == 0, it will be automatically set to n_threads. * to explicitly set n_cpus, just pass the value. This is primarily intended for * the purpose of explicitly constraining rendering parallelization to less than n_threads, -- cgit v1.2.1