intel/compiler: Allow for varying subgroup sizes

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>

diff --git a/src/intel/compiler/brw_compiler.h b/src/intel/compiler/brw_compiler.h
index ba95df5b7dc887f7caf1beeb33aff943f04ee178..8b62e67e2d7cfeda4ede4fa9ad73326e6cd71b09 100644 (file)
--- a/src/intel/compiler/brw_compiler.h
+++ b/src/intel/compiler/brw_compiler.h
@@ -208,6 +208,7 @@ enum PACKED brw_subgroup_size_type
 {
    BRW_SUBGROUP_SIZE_API_CONSTANT,  /**< Vulkan behavior */
    BRW_SUBGROUP_SIZE_UNIFORM,       /**< OpenGL behavior */
+   BRW_SUBGROUP_SIZE_VARYING,       /**< VK_EXT_subgroup_size_control */
 };
 
 struct brw_base_prog_key {

diff --git a/src/intel/compiler/brw_fs_nir.cpp b/src/intel/compiler/brw_fs_nir.cpp
index e5a927431c4e01f28a7b3118b3c6cda85429699a..6ea012c74b62361165b2efaba04a93a830ddcebd 100644 (file)
--- a/src/intel/compiler/brw_fs_nir.cpp
+++ b/src/intel/compiler/brw_fs_nir.cpp
@@ -4916,6 +4916,14 @@ fs_visitor::nir_emit_intrinsic(const fs_builder &bld, nir_intrinsic_instr *instr
       break;
    }
 
+   case nir_intrinsic_load_subgroup_size:
+      /* This should only happen for fragment shaders because every other case
+       * is lowered in NIR so we can optimize on it.
+       */
+      assert(stage == MESA_SHADER_FRAGMENT);
+      bld.MOV(retype(dest, BRW_REGISTER_TYPE_D), brw_imm_d(dispatch_width));
+      break;
+
    case nir_intrinsic_load_subgroup_invocation:
       bld.MOV(retype(dest, BRW_REGISTER_TYPE_D),
               nir_system_values[SYSTEM_VALUE_SUBGROUP_INVOCATION]);

diff --git a/src/intel/compiler/brw_nir.c b/src/intel/compiler/brw_nir.c
index 9e4b33c8b49db6de4fb35ce2569491f1ea928d24..3f90ab5a2baeda3da91aa731c867792d40df7c9e 100644 (file)
--- a/src/intel/compiler/brw_nir.c
+++ b/src/intel/compiler/brw_nir.c
@@ -967,7 +967,8 @@ brw_nir_apply_sampler_key(nir_shader *nir,
 }
 
 static unsigned
-get_subgroup_size(const struct brw_base_prog_key *key,
+get_subgroup_size(gl_shader_stage stage,
+                  const struct brw_base_prog_key *key,
                   unsigned max_subgroup_size)
 {
    switch (key->subgroup_size_type) {
@@ -985,6 +986,19 @@ get_subgroup_size(const struct brw_base_prog_key *key,
        * to be uniform across invocations.
        */
       return max_subgroup_size;
+
+   case BRW_SUBGROUP_SIZE_VARYING:
+      /* The subgroup size is allowed to be fully varying.  For geometry
+       * stages, we know it's always 8 which is max_subgroup_size so we can
+       * return that.  For compute, brw_nir_apply_key is called once per
+       * dispatch-width so max_subgroup_size is the real subgroup size.
+       *
+       * For fragment, we return 0 and let it fall through to the back-end
+       * compiler.  This means we can't optimize based on subgroup size but
+       * that's a risk the client took when it asked for a varying subgroup
+       * size.
+       */
+      return stage == MESA_SHADER_FRAGMENT ? 0 : max_subgroup_size;
    }
 
    unreachable("Invalid subgroup size type");
@@ -1002,7 +1016,8 @@ brw_nir_apply_key(nir_shader *nir,
    OPT(brw_nir_apply_sampler_key, compiler, &key->tex);
 
    const nir_lower_subgroups_options subgroups_options = {
-      .subgroup_size = get_subgroup_size(key, max_subgroup_size),
+      .subgroup_size = get_subgroup_size(nir->info.stage, key,
+                                         max_subgroup_size),
       .ballot_bit_size = 32,
       .lower_subgroup_masks = true,
    };