On 04/18/2017 03:39 PM, Jason Ekstrand wrote: > On Mon, Apr 17, 2017 at 4:49 PM, Edward O'Callaghan > <funfunc...@folklore1984.net <mailto:funfunc...@folklore1984.net>> wrote: > > > > On 04/18/2017 01:13 AM, Jason Ekstrand wrote: > > I don't see anything in here that actually implements parallel shader > > compilation. Does radeon already spawn back-end threads to compile > As far as I am aware that is exactly right. > > > If that's the case, then I didn't see anywhere that you actually hooked > the maximum number of threads value to the radeon back-end. The > extension does two things: > > 1) Allow for asynchronous reading of compile status > 2) Allow the user to specify the maximum number of threads for the > driver to use for shader compiles. > > Your implementation does neither of these things.
Correct, actually the most important which is to make glCompile and glLink to be async not just reading the status, that is what I am working on locally. The spec doesn't actually put a hard requirement on the number of threads to be exactly used, it is purely a hint and implementation dependent. That said, it is indeed intended that Compile/Link be threaded so that is the 90% left yes. Ilia helped me answer my frontend XML questions that I was stuck with. Thanks for the feedback guys, > > > > shaders or are we just lying about it? Even if it does, the front-end > > tends to take significant quantities of time so it'd be nice to > > parallelize that. > Indeed the frontend part I believe is where the remaining work is to be > done. > > > Right. 90% of the work is left to be done. > > --Jason > > > > > > On Mon, Apr 17, 2017 at 5:07 AM, Edward O'Callaghan > > <funfunc...@folklore1984.net <mailto:funfunc...@folklore1984.net> > <mailto:funfunc...@folklore1984.net > <mailto:funfunc...@folklore1984.net>>> wrote: > > > > Signed-off-by: Edward O'Callaghan <funfunc...@folklore1984.net > <mailto:funfunc...@folklore1984.net> > > <mailto:funfunc...@folklore1984.net > <mailto:funfunc...@folklore1984.net>>> > > --- > > src/mesa/main/get.c | 4 ++++ > > src/mesa/main/mtypes.h | 6 ++++++ > > src/mesa/main/shaderapi.c | 20 ++++++++++++++++++++ > > src/mesa/main/shaderapi.h | 3 +++ > > src/mesa/main/shared.c | 3 +++ > > src/mesa/state_tracker/st_extensions.c | 4 ++++ > > 6 files changed, 40 insertions(+) > > > > diff --git a/src/mesa/main/get.c b/src/mesa/main/get.c > > index cf3ee63..91f21ce 100644 > > --- a/src/mesa/main/get.c > > +++ b/src/mesa/main/get.c > > @@ -2332,6 +2332,10 @@ find_value_indexed(const char *func, GLenum > > pname, GLuint index, union value *v) > > return TYPE_INT; > > } > > > > + case GL_MAX_SHADER_COMPILER_THREADS_ARB: > > + v->value_int = ctx->Shared->MaxCompileThreads; > > + return TYPE_INT; > > + > > case GL_MAX_COMPUTE_WORK_GROUP_COUNT: > > if (!_mesa_has_compute_shaders(ctx)) > > goto invalid_enum; > > diff --git a/src/mesa/main/mtypes.h b/src/mesa/main/mtypes.h > > index 4a76de5..33cf33d 100644 > > --- a/src/mesa/main/mtypes.h > > +++ b/src/mesa/main/mtypes.h > > @@ -3227,6 +3227,9 @@ struct gl_shared_state > > * Once this field becomes true, it is never reset to false. > > */ > > bool ShareGroupReset; > > + > > + /* GL_ARB_parallel_shader_compile */ > > + int MaxCompileThreads; > > > > > > This seems to only be initialized by st/mesa. Shoule we initialize it > > to 1 by default for everyone else? > > > > > > }; > > > > > > @@ -3880,6 +3883,9 @@ struct gl_constants > > GLuint MaxImageSamples; > > GLuint MaxCombinedImageUniforms; > > > > + /** GL_ARB_parallel_shader_compile */ > > + GLuint MaxShaderCompilerThreads; > > + > > /** GL_ARB_compute_shader */ > > GLuint MaxComputeWorkGroupCount[3]; /* Array of x, y, z > > dimensions */ > > GLuint MaxComputeWorkGroupSize[3]; /* Array of x, y, z > dimensions */ > > diff --git a/src/mesa/main/shaderapi.c b/src/mesa/main/shaderapi.c > > index 187475f..051390a 100644 > > --- a/src/mesa/main/shaderapi.c > > +++ b/src/mesa/main/shaderapi.c > > @@ -631,6 +631,7 @@ get_programiv(struct gl_context *ctx, GLuint > > program, GLenum pname, > > case GL_DELETE_STATUS: > > *params = shProg->DeletePending; > > return; > > + case GL_COMPLETION_STATUS_ARB: > > case GL_LINK_STATUS: > > *params = shProg->data->LinkStatus ? GL_TRUE : GL_FALSE; > > return; > > @@ -898,6 +899,7 @@ get_shaderiv(struct gl_context *ctx, GLuint > > name, GLenum pname, GLint *params) > > case GL_DELETE_STATUS: > > *params = shader->DeletePending; > > break; > > + case GL_COMPLETION_STATUS_ARB: > > case GL_COMPILE_STATUS: > > *params = shader->CompileStatus ? GL_TRUE : GL_FALSE; > > break; > > @@ -2234,6 +2236,24 @@ _mesa_copy_linked_program_data(const struct > > gl_shader_program *src, > > } > > > > /** > > + * ARB_parallel_shader_compile > > + */ > > +GLvoid GLAPIENTRY > > +_mesa_MaxShaderCompilerThreadsARB(GLuint count) > > +{ > > + GET_CURRENT_CONTEXT(ctx); > > + > > + /** > > + * Additions to Chapter 7 "Programs and Shaders": > > + * a <count> of 0xFFFFFFFF requests an implementation-specific > > maximum. > > + */ > > + if (count == 0xFFFFFFFF || count > > > ctx->Const.MaxShaderCompilerThreads) > > + count = ctx->Const.MaxShaderCompilerThreads; > > > > > > Give that it's a 32-bit unsigned integer, 0xFFFFFFFF > > > ctx->Const.MaxShaderCompilerThreads so we don't need the extra case. > > > > > > + > > + ctx->Shared->MaxCompileThreads = count; > > +} > > + > > +/** > > * ARB_separate_shader_objects: Compile & Link Program > > */ > > GLuint GLAPIENTRY > > diff --git a/src/mesa/main/shaderapi.h b/src/mesa/main/shaderapi.h > > index 99b4fe8..1ec287a 100644 > > --- a/src/mesa/main/shaderapi.h > > +++ b/src/mesa/main/shaderapi.h > > @@ -225,6 +225,9 @@ _mesa_copy_linked_program_data(const struct > > gl_shader_program *src, > > extern bool > > _mesa_validate_shader_target(const struct gl_context *ctx, GLenum > > type); > > > > +/* ARB_parallel_shader_compile */ > > +extern GLvoid GLAPIENTRY > > +_mesa_MaxShaderCompilerThreadsARB(GLuint count); > > > > /* GL_ARB_separate_shader_objects */ > > extern GLuint GLAPIENTRY > > diff --git a/src/mesa/main/shared.c b/src/mesa/main/shared.c > > index 5344812..586502e 100644 > > --- a/src/mesa/main/shared.c > > +++ b/src/mesa/main/shared.c > > @@ -126,6 +126,9 @@ _mesa_alloc_shared_state(struct gl_context > *ctx) > > shared->SyncObjects = _mesa_set_create(NULL, > _mesa_hash_pointer, > > > _mesa_key_pointer_equal); > > > > + /* GL_ARB_parallel_shader_compile */ > > + shared->MaxCompileThreads = 0xFFFFFFFF; > > + > > return shared; > > } > > > > diff --git a/src/mesa/state_tracker/st_extensions.c > > b/src/mesa/state_tracker/st_extensions.c > > index 1df2ba7..61391f8 100644 > > --- a/src/mesa/state_tracker/st_extensions.c > > +++ b/src/mesa/state_tracker/st_extensions.c > > @@ -35,6 +35,7 @@ > > #include "pipe/p_defines.h" > > #include "pipe/p_screen.h" > > #include "util/u_math.h" > > +#include "util/u_cpu_detect.h" > > > > #include "st_context.h" > > #include "st_debug.h" > > @@ -897,6 +898,9 @@ void st_init_extensions(struct pipe_screen > *screen, > > if (consts->GLSLVersion >= 410) > > extensions->ARB_shader_precision = GL_TRUE; > > > > + /* GL_ARB_parallel_shader_compile */ > > + consts->MaxShaderCompilerThreads = util_cpu_caps.nr_cpus > 1 ? > > util_cpu_caps.nr_cpus : 0; > > + > > /* This extension needs full OpenGL 3.2, but we don't know > if that's > > * supported at this point. Only check the GLSL version. */ > > if (consts->GLSLVersion >= 150 && > > -- > > 2.9.3 > > > > _______________________________________________ > > mesa-dev mailing list > > mesa-dev@lists.freedesktop.org > <mailto:mesa-dev@lists.freedesktop.org> > <mailto:mesa-dev@lists.freedesktop.org > <mailto:mesa-dev@lists.freedesktop.org>> > > https://lists.freedesktop.org/mailman/listinfo/mesa-dev > <https://lists.freedesktop.org/mailman/listinfo/mesa-dev> > > <https://lists.freedesktop.org/mailman/listinfo/mesa-dev > <https://lists.freedesktop.org/mailman/listinfo/mesa-dev>> > > > > > >
signature.asc
Description: OpenPGP digital signature
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
