radv: use radv_optimize_nir() less in radv_link_shaders()

fossil-db (Navi):
Totals from 11 (0.01% of 137413) affected shaders:
CodeSize: 99372 -> 99480 (+0.11%)
Instrs: 19119 -> 19110 (-0.05%)
Cycles: 222144 -> 222000 (-0.06%)

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6891>

diff --git a/src/amd/vulkan/radv_pipeline.c b/src/amd/vulkan/radv_pipeline.c
index 7bba42e..2c3c8cd 100644 (file)
--- a/src/amd/vulkan/radv_pipeline.c
+++ b/src/amd/vulkan/radv_pipeline.c
@@ -2226,20 +2226,47 @@ radv_link_shaders(struct radv_pipeline *pipeline, nir_shader **shaders,
                        if (ordered_shaders[i]->info.stage != last)
                                mask = mask | nir_var_shader_out;
 
-                       nir_lower_io_to_scalar_early(ordered_shaders[i], mask);
+                       if (nir_lower_io_to_scalar_early(ordered_shaders[i], mask)) {
+                               /* Optimize the new vector code and then remove dead vars */
+                               nir_copy_prop(ordered_shaders[i]);
+                               nir_opt_shrink_vectors(ordered_shaders[i]);
+
+                               if (ordered_shaders[i]->info.stage != last) {
+                                       /* Optimize swizzled movs of load_const for
+                                        * nir_link_opt_varyings's constant propagation
+                                        */
+                                       nir_opt_constant_folding(ordered_shaders[i]);
+                                       /* For nir_link_opt_varyings's duplicate input opt */
+                                       nir_opt_cse(ordered_shaders[i]);
+                               }
+
+                               /* Run copy-propagation to help remove dead
+                                * output variables (some shaders have useless
+                                * copies to/from an output), so compaction
+                                * later will be more effective.
+                                *
+                                * This will have been done earlier but it might
+                                * not have worked because the outputs were vector.
+                                */
+                               if (ordered_shaders[i]->info.stage == MESA_SHADER_TESS_CTRL)
+                                       nir_opt_copy_prop_vars(ordered_shaders[i]);
+
+                               nir_opt_dce(ordered_shaders[i]);
+                               nir_remove_dead_variables(ordered_shaders[i],
+                                                         nir_var_function_temp | nir_var_shader_in | nir_var_shader_out, NULL);
+                       }
                }
        }
 
-       for (int i = 0; i < shader_count; ++i)
-               radv_optimize_nir(ordered_shaders[i], optimize_conservatively, false);
-
        for (int i = 1; !optimize_conservatively && (i < shader_count); ++i)  {
                nir_lower_io_arrays_to_elements(ordered_shaders[i],
                                                ordered_shaders[i - 1]);
 
-               if (nir_link_opt_varyings(ordered_shaders[i],
-                                         ordered_shaders[i - 1]))
-                       radv_optimize_nir(ordered_shaders[i - 1], false, false);
+               if (nir_link_opt_varyings(ordered_shaders[i], ordered_shaders[i - 1])) {
+                       nir_opt_constant_folding(ordered_shaders[i - 1]);
+                       nir_opt_algebraic(ordered_shaders[i - 1]);
+                       nir_opt_dce(ordered_shaders[i - 1]);
+               }
 
                nir_remove_dead_variables(ordered_shaders[i],
                                          nir_var_shader_out, NULL);
@@ -2256,16 +2283,20 @@ radv_link_shaders(struct radv_pipeline *pipeline, nir_shader **shaders,
                        if (nir_lower_global_vars_to_local(ordered_shaders[i])) {
                                ac_lower_indirect_derefs(ordered_shaders[i],
                                                         pipeline->device->physical_device->rad_info.chip_class);
+                               /* remove dead writes, which can remove input loads */
+                               nir_lower_vars_to_ssa(ordered_shaders[i]);
+                               nir_opt_dce(ordered_shaders[i]);
                        }
-                       radv_optimize_nir(ordered_shaders[i], false, false);
 
                        if (nir_lower_global_vars_to_local(ordered_shaders[i - 1])) {
                                ac_lower_indirect_derefs(ordered_shaders[i - 1],
                                                         pipeline->device->physical_device->rad_info.chip_class);
                        }
-                       radv_optimize_nir(ordered_shaders[i - 1], false, false);
                }
        }
+
+       for (int i = 0; i < shader_count; ++i)
+               radv_optimize_nir(ordered_shaders[i], optimize_conservatively, false);
 }
 
 static void

diff --git a/src/amd/vulkan/radv_shader.c b/src/amd/vulkan/radv_shader.c
index bc52b95..a80b3a9 100644 (file)
--- a/src/amd/vulkan/radv_shader.c
+++ b/src/amd/vulkan/radv_shader.c
@@ -649,7 +649,12 @@ radv_shader_compile_to_nir(struct radv_device *device,
         * bloat the instruction count of the loop and cause it to be
         * considered too large for unrolling.
         */
-       ac_lower_indirect_derefs(nir, device->physical_device->rad_info.chip_class);
+       if (ac_lower_indirect_derefs(nir, device->physical_device->rad_info.chip_class) &&
+           !(flags & VK_PIPELINE_CREATE_DISABLE_OPTIMIZATION_BIT) &&
+           nir->info.stage != MESA_SHADER_COMPUTE) {
+               /* Optimize the lowered code before the linking optimizations. */
+               radv_optimize_nir(nir, false, false);
+       }
 
        return nir;
 }