glsl: Make copy propagation not panic when it sees an intrinsic.

A number of games have large arrays of constants, which we promote to
uniforms.  This introduces copies from the uniform array to the original
temporary array.  Normally, copy propagation eliminates those copies,
making everything refer to the uniform array directly.

A number of shaders in "Deus Ex: Mankind Divided" recently exposed a
limitation of copy propagation - if we had any intrinsics (i.e. image
access in a compute shader), we weren't able to get rid of these copies.

That meant that any variable indexing remained on the temporary array
rather being moved to the uniform array.  i965's scalar backend
currently doesn't support indirect addressing of temporary arrays,
which meant lowering it to if-ladders.  This was horrible.

According to Marek, on radeonsi/GCN, "F1 2015" uses 64% less
spilled-temp-array memory.

On i965/Skylake:

total instructions in shared programs: 13362954 -> 13329878 (-0.25%)
instructions in affected programs: 43745 -> 10669 (-75.61%)
helped: 12
HURT: 0

total cycles in shared programs: 248081010 -> 245949178 (-0.86%)
cycles in affected programs: 4597930 -> 2466098 (-46.37%)
helped: 12
HURT: 0

total spills in shared programs: 9493 -> 9507 (0.15%)
spills in affected programs: 25 -> 39 (56.00%)
helped: 0
HURT: 1

total fills in shared programs: 12127 -> 12197 (0.58%)
fills in affected programs: 110 -> 180 (63.64%)
helped: 0
HURT: 1

Helps Deus Ex: Mankind Divided.   The one shader with hurt spills/fills
is from Tomb Raider at Ultra settings, but that same shader has a
-39.55% reduction in instructions and -14.09% reduction in cycle counts,
so it seems like a win there as well.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>

diff --git a/src/compiler/glsl/opt_copy_propagation.cpp b/src/compiler/glsl/opt_copy_propagation.cpp
index 247c4988ed3d03502c926dc4390274985b6ed965..2240421a2a5dbc795fa0945d9428d4a1f4f3357c 100644 (file)
--- a/src/compiler/glsl/opt_copy_propagation.cpp
+++ b/src/compiler/glsl/opt_copy_propagation.cpp
@@ -186,11 +186,34 @@ ir_copy_propagation_visitor::visit_enter(ir_call *ir)
       }
    }
 
-   /* Since we're unlinked, we don't (necessarily) know the side effects of
-    * this call.  So kill all copies.
+   /* Since this pass can run when unlinked, we don't (necessarily) know
+    * the side effects of calls.  (When linked, most calls are inlined
+    * anyway, so it doesn't matter much.)
+    *
+    * One place where this does matter is IR intrinsics.  They're never
+    * inlined.  We also know what they do - while some have side effects
+    * (such as image writes), none edit random global variables.  So we
+    * can assume they're side-effect free (other than the return value
+    * and out parameters).
     */
-   _mesa_hash_table_clear(acp, NULL);
-   this->killed_all = true;
+   if (!ir->callee->is_intrinsic()) {
+      _mesa_hash_table_clear(acp, NULL);
+      this->killed_all = true;
+   } else {
+      if (ir->return_deref)
+         kill(ir->return_deref->var);
+
+      foreach_two_lists(formal_node, &ir->callee->parameters,
+                        actual_node, &ir->actual_parameters) {
+         ir_variable *sig_param = (ir_variable *) formal_node;
+         if (sig_param->data.mode == ir_var_function_out ||
+             sig_param->data.mode == ir_var_function_inout) {
+            ir_rvalue *ir = (ir_rvalue *) actual_node;
+            ir_variable *var = ir->variable_referenced();
+            kill(var);
+         }
+      }
+   }
 
    return visit_continue_with_parent;
 }