intel/compiler: Track patch count threshold

Return the number of patches to accumulate before an 8_PATCH mode thread
is launched.

v2: (Kenneth Graunke)
- Track patch count threshold instead of input control points.

Signed-off-by: Sagar Ghuge <[email protected]>
Reviewed-by: Kenneth Graunke <[email protected]>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3563>

2 files changed, 33 insertions, 0 deletions

diff --git a/src/intel/compiler/brw_compiler.h b/src/intel/compiler/brw_compiler.h
index 181f7e5f67c..2048cfaafcc 100644
--- a/src/intel/compiler/brw_compiler.h
+++ b/src/intel/compiler/brw_compiler.h
@@ -1136,6 +1136,9 @@ struct brw_tcs_prog_data
 
    /** Number vertices in output patch */
    int instances;
+
+   /** Track patch count threshold */
+   int patch_count_threshold;
 };
 
 
diff --git a/src/intel/compiler/brw_vec4_tcs.cpp b/src/intel/compiler/brw_vec4_tcs.cpp
index 1cba9a9a4f8..fcefe395f2d 100644
--- a/src/intel/compiler/brw_vec4_tcs.cpp
+++ b/src/intel/compiler/brw_vec4_tcs.cpp
@@ -323,6 +323,34 @@ vec4_tcs_visitor::nir_emit_intrinsic(nir_intrinsic_instr *instr)
    }
 }
 
+/**
+ * Return the number of patches to accumulate before an 8_PATCH mode thread is
+ * launched.  In cases with a large number of input control points and a large
+ * amount of VS outputs, the VS URB space needed to store an entire 8 patches
+ * worth of data can be prohibitive, so it can be beneficial to launch threads
+ * early.
+ *
+ * See the 3DSTATE_HS::Patch Count Threshold documentation for the recommended
+ * values.  Note that 0 means to "disable" early dispatch, meaning to wait for
+ * a full 8 patches as normal.
+ */
+static int
+get_patch_count_threshold(int input_control_points)
+{
+   if (input_control_points <= 4)
+      return 0;
+   else if (input_control_points <= 6)
+      return 5;
+   else if (input_control_points <= 8)
+      return 4;
+   else if (input_control_points <= 10)
+      return 3;
+   else if (input_control_points <= 14)
+      return 2;
+
+   /* Return patch count 1 for PATCHLIST_15 - PATCHLIST_32 */
+   return 1;
+}
 
 extern "C" const unsigned *
 brw_compile_tcs(const struct brw_compiler *compiler,
@@ -362,6 +390,8 @@ brw_compile_tcs(const struct brw_compiler *compiler,
    bool has_primitive_id =
       nir->info.system_values_read & (1 << SYSTEM_VALUE_PRIMITIVE_ID);
 
+   prog_data->patch_count_threshold = get_patch_count_threshold(key->input_vertices);
+
    if (compiler->use_tcs_8_patch &&
        nir->info.tess.tcs_vertices_out <= (devinfo->gen >= 12 ? 32 : 16) &&
        2 + has_primitive_id + key->input_vertices <= 31) {