[Mesa-dev] [PATCH v3 29/48] intel/cs: Rework the way thread local ID is handled

Iago Toral itoral at igalia.com
Fri Oct 27 09:11:14 UTC 2017


On Wed, 2017-10-25 at 16:26 -0700, Jason Ekstrand wrote:
> Previously, brw_nir_lower_intrinsics added the param and then emitted
> a
> load_uniform intrinsic to load it directly.  This commit switches
> things
> over to use a specific NIR intrinsic for the thread id.  The one
> thing I
> don't like about this approach is that we have to copy
> thread_local_id
> over to the new visitor in import_uniforms.

It is not clear to me why you are doing this... why do you like this
better?

> ---
>  src/compiler/nir/nir_intrinsics.h                |  3 ++
>  src/intel/compiler/brw_fs.cpp                    |  4 +-
>  src/intel/compiler/brw_fs.h                      |  1 +
>  src/intel/compiler/brw_fs_nir.cpp                | 14 +++++++
>  src/intel/compiler/brw_nir.h                     |  3 +-
>  src/intel/compiler/brw_nir_lower_cs_intrinsics.c | 53 +++++---------
> ----------
>  6 files changed, 32 insertions(+), 46 deletions(-)
> 
> diff --git a/src/compiler/nir/nir_intrinsics.h
> b/src/compiler/nir/nir_intrinsics.h
> index cefd18b..47022dd 100644
> --- a/src/compiler/nir/nir_intrinsics.h
> +++ b/src/compiler/nir/nir_intrinsics.h
> @@ -364,6 +364,9 @@ SYSTEM_VALUE(blend_const_color_a_float, 1, 0, xx,
> xx, xx)
>  SYSTEM_VALUE(blend_const_color_rgba8888_unorm, 1, 0, xx, xx, xx)
>  SYSTEM_VALUE(blend_const_color_aaaa8888_unorm, 1, 0, xx, xx, xx)
>  
> +/* Intel specific system values */
> +SYSTEM_VALUE(intel_thread_local_id, 1, 0, xx, xx, xx)
> +
>  /**
>   * Barycentric coordinate intrinsics.
>   *
> diff --git a/src/intel/compiler/brw_fs.cpp
> b/src/intel/compiler/brw_fs.cpp
> index 2acd838..c0d4c05 100644
> --- a/src/intel/compiler/brw_fs.cpp
> +++ b/src/intel/compiler/brw_fs.cpp
> @@ -996,6 +996,7 @@ fs_visitor::import_uniforms(fs_visitor *v)
>     this->push_constant_loc = v->push_constant_loc;
>     this->pull_constant_loc = v->pull_constant_loc;
>     this->uniforms = v->uniforms;
> +   this->thread_local_id = v->thread_local_id;
>  }
>  
>  void
> @@ -6781,8 +6782,7 @@ brw_compile_cs(const struct brw_compiler
> *compiler, void *log_data,
>  {
>     nir_shader *shader = nir_shader_clone(mem_ctx, src_shader);
>     shader = brw_nir_apply_sampler_key(shader, compiler, &key->tex,
> true);
> -
> -   brw_nir_lower_cs_intrinsics(shader, prog_data);
> +   brw_nir_lower_cs_intrinsics(shader);
>     shader = brw_postprocess_nir(shader, compiler, true);
>  
>     prog_data->local_size[0] = shader->info.cs.local_size[0];
> diff --git a/src/intel/compiler/brw_fs.h
> b/src/intel/compiler/brw_fs.h
> index da32593..f51a4d8 100644
> --- a/src/intel/compiler/brw_fs.h
> +++ b/src/intel/compiler/brw_fs.h
> @@ -315,6 +315,7 @@ public:
>      */
>     int *push_constant_loc;
>  
> +   fs_reg thread_local_id;
>     fs_reg frag_depth;
>     fs_reg frag_stencil;
>     fs_reg sample_mask;
> diff --git a/src/intel/compiler/brw_fs_nir.cpp
> b/src/intel/compiler/brw_fs_nir.cpp
> index 05efee3..fdc6fc6 100644
> --- a/src/intel/compiler/brw_fs_nir.cpp
> +++ b/src/intel/compiler/brw_fs_nir.cpp
> @@ -88,6 +88,16 @@ fs_visitor::nir_setup_uniforms()
>     }
>  
>     uniforms = nir->num_uniforms / 4;
> +
> +   if (stage == MESA_SHADER_COMPUTE) {
> +      /* Add a uniform for the thread local id.  It must be the last
> uniform
> +       * on the list.
> +       */
> +      assert(uniforms == prog_data->nr_params);
> +      uint32_t *param = brw_stage_prog_data_add_params(prog_data,
> 1);
> +      *param = BRW_PARAM_BUILTIN_THREAD_LOCAL_ID;
> +      thread_local_id = fs_reg(UNIFORM, uniforms++,
> BRW_REGISTER_TYPE_UD);
> +   }
>  }
>  
>  static bool
> @@ -3409,6 +3419,10 @@ fs_visitor::nir_emit_cs_intrinsic(const
> fs_builder &bld,
>        cs_prog_data->uses_barrier = true;
>        break;
>  
> +   case nir_intrinsic_load_intel_thread_local_id:
> +      bld.MOV(retype(dest, BRW_REGISTER_TYPE_UD), thread_local_id);
> +      break;
> +
>     case nir_intrinsic_load_local_invocation_id:
>     case nir_intrinsic_load_work_group_id: {
>        gl_system_value sv = nir_system_value_from_intrinsic(instr-
> >intrinsic);
> diff --git a/src/intel/compiler/brw_nir.h
> b/src/intel/compiler/brw_nir.h
> index 1493b74..3e40712 100644
> --- a/src/intel/compiler/brw_nir.h
> +++ b/src/intel/compiler/brw_nir.h
> @@ -95,8 +95,7 @@ void brw_nir_analyze_boolean_resolves(nir_shader
> *nir);
>  nir_shader *brw_preprocess_nir(const struct brw_compiler *compiler,
>                                 nir_shader *nir);
>  
> -bool brw_nir_lower_cs_intrinsics(nir_shader *nir,
> -                                 struct brw_cs_prog_data
> *prog_data);
> +bool brw_nir_lower_cs_intrinsics(nir_shader *nir);
>  void brw_nir_lower_vs_inputs(nir_shader *nir,
>                               bool use_legacy_snorm_formula,
>                               const uint8_t *vs_attrib_wa_flags);
> diff --git a/src/intel/compiler/brw_nir_lower_cs_intrinsics.c
> b/src/intel/compiler/brw_nir_lower_cs_intrinsics.c
> index d277276..07d2dcc 100644
> --- a/src/intel/compiler/brw_nir_lower_cs_intrinsics.c
> +++ b/src/intel/compiler/brw_nir_lower_cs_intrinsics.c
> @@ -26,47 +26,12 @@
>  
>  struct lower_intrinsics_state {
>     nir_shader *nir;
> -   struct brw_cs_prog_data *prog_data;
>     nir_function_impl *impl;
>     bool progress;
>     nir_builder builder;
> -   int thread_local_id_index;
> +   unsigned local_workgroup_size;
>  };
>  
> -static nir_ssa_def *
> -read_thread_local_id(struct lower_intrinsics_state *state)
> -{
> -   struct brw_cs_prog_data *prog_data = state->prog_data;
> -   nir_builder *b = &state->builder;
> -   nir_shader *nir = state->nir;
> -   const unsigned *sizes = nir->info.cs.local_size;
> -   const unsigned group_size = sizes[0] * sizes[1] * sizes[2];
> -
> -   /* Some programs have local_size dimensions so small that the
> thread local
> -    * ID will always be 0.
> -    */
> -   if (group_size <= 8)
> -      return nir_imm_int(b, 0);
> -
> -   if (state->thread_local_id_index == -1) {
> -      state->thread_local_id_index = prog_data->base.nr_params;
> -      uint32_t *param = brw_stage_prog_data_add_params(&prog_data-
> >base, 1);
> -      *param = BRW_PARAM_BUILTIN_THREAD_LOCAL_ID;
> -      nir->num_uniforms += 4;
> -   }
> -   unsigned id_index = state->thread_local_id_index;
> -
> -   nir_intrinsic_instr *load =
> -      nir_intrinsic_instr_create(nir, nir_intrinsic_load_uniform);
> -   load->num_components = 1;
> -   load->src[0] = nir_src_for_ssa(nir_imm_int(b, 0));
> -   nir_ssa_dest_init(&load->instr, &load->dest, 1, 32, NULL);
> -   nir_intrinsic_set_base(load, id_index * sizeof(uint32_t));
> -   nir_intrinsic_set_range(load, sizeof(uint32_t));
> -   nir_builder_instr_insert(b, &load->instr);
> -   return &load->dest.ssa;
> -}
> -
>  static bool
>  lower_cs_intrinsics_convert_block(struct lower_intrinsics_state
> *state,
>                                    nir_block *block)
> @@ -91,7 +56,12 @@ lower_cs_intrinsics_convert_block(struct
> lower_intrinsics_state *state,
>            *    gl_LocalInvocationIndex =
>            *       cs_thread_local_id + subgroup_invocation;
>            */
> -         nir_ssa_def *thread_local_id = read_thread_local_id(state);
> +         nir_ssa_def *thread_local_id;
> +         if (state->local_workgroup_size <= 8)
> +            thread_local_id = nir_imm_int(b, 0);
> +         else
> +            thread_local_id = nir_load_intel_thread_local_id(b);
> +
>           nir_ssa_def *channel = nir_load_subgroup_invocation(b);
>           sysval = nir_iadd(b, channel, thread_local_id);
>           break;
> @@ -157,8 +127,7 @@ lower_cs_intrinsics_convert_impl(struct
> lower_intrinsics_state *state)
>  }
>  
>  bool
> -brw_nir_lower_cs_intrinsics(nir_shader *nir,
> -                            struct brw_cs_prog_data *prog_data)
> +brw_nir_lower_cs_intrinsics(nir_shader *nir)
>  {
>     assert(nir->info.stage == MESA_SHADER_COMPUTE);
>  
> @@ -166,9 +135,9 @@ brw_nir_lower_cs_intrinsics(nir_shader *nir,
>     struct lower_intrinsics_state state;
>     memset(&state, 0, sizeof(state));
>     state.nir = nir;
> -   state.prog_data = prog_data;
> -
> -   state.thread_local_id_index = -1;
> +   state.local_workgroup_size = nir->info.cs.local_size[0] *
> +                                nir->info.cs.local_size[1] *
> +                                nir->info.cs.local_size[2];
>  
>     do {
>        state.progress = false;


More information about the mesa-dev mailing list