intel/fs: Combine adjacent memory barriers

This will avoid generating multiple identical fences in a row.

For Gen11+ we have multiple types of fences (affecting different
variable modes), but is still better to combine them in a single
scoped barrier so that the translation to backend IR have the option
of dispatching both fences in parallel.

This will clean up redundant barriers from various
dEQP-VK.memory_model.* tests.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Tested-by: Marge Bot <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3224>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3224>

diff --git a/src/intel/compiler/brw_nir.c b/src/intel/compiler/brw_nir.c
index 6c6e72208acebb4f27e96cc85fec0b9b78eb6385..9e8ae3b52a1fa4e2824903285640b8e1cf81ce12 100644 (file)
--- a/src/intel/compiler/brw_nir.c
+++ b/src/intel/compiler/brw_nir.c
@@ -824,6 +824,26 @@ brw_nir_link_shaders(const struct brw_compiler *compiler,
    }
 }
 
+static
+bool combine_all_barriers(nir_intrinsic_instr *a,
+                          nir_intrinsic_instr *b,
+                          void *data)
+{
+   /* Translation to backend IR will get rid of modes we don't care about, so
+    * no harm in always combining them.
+    *
+    * TODO: While HW has only ACQUIRE|RELEASE fences, we could improve the
+    * scheduling so that it can take advantage of the different semantics.
+    */
+   nir_intrinsic_set_memory_modes(a, nir_intrinsic_memory_modes(a) |
+                                     nir_intrinsic_memory_modes(b));
+   nir_intrinsic_set_memory_semantics(a, nir_intrinsic_memory_semantics(a) |
+                                         nir_intrinsic_memory_semantics(b));
+   nir_intrinsic_set_memory_scope(a, MAX2(nir_intrinsic_memory_scope(a),
+                                          nir_intrinsic_memory_scope(b)));
+   return true;
+}
+
 /* Prepare the given shader for codegen
  *
  * This function is intended to be called right before going into the actual
@@ -843,6 +863,8 @@ brw_postprocess_nir(nir_shader *nir, const struct brw_compiler *compiler,
 
    OPT(brw_nir_lower_mem_access_bit_sizes, devinfo);
 
+   OPT(nir_opt_combine_memory_barriers, combine_all_barriers, NULL);
+
    do {
       progress = false;
       OPT(nir_opt_algebraic_before_ffma);