glsl: don't lower variable indexing on non-patch tessellation inputs/outputs

There is no way to lower them, because the array sizes are unknown
at compile time.

Based on a patch from: Fabian Bieler <fabianbieler@fastmail.fm>

v2: add comments

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Dave Airlie <airlied@redhat.com>

diff --git a/src/glsl/ir_optimization.h b/src/glsl/ir_optimization.h
index 688a5e1..a174c96 100644 (file)
--- a/src/glsl/ir_optimization.h
+++ b/src/glsl/ir_optimization.h
@@ -114,8 +114,9 @@ bool lower_discard(exec_list *instructions);
 void lower_discard_flow(exec_list *instructions);
 bool lower_instructions(exec_list *instructions, unsigned what_to_lower);
 bool lower_noise(exec_list *instructions);
-bool lower_variable_index_to_cond_assign(exec_list *instructions,
-    bool lower_input, bool lower_output, bool lower_temp, bool lower_uniform);
+bool lower_variable_index_to_cond_assign(gl_shader_stage stage,
+    exec_list *instructions, bool lower_input, bool lower_output,
+    bool lower_temp, bool lower_uniform);
 bool lower_quadop_vector(exec_list *instructions, bool dont_lower_swz);
 bool lower_const_arrays_to_uniforms(exec_list *instructions);
 bool lower_clip_distance(gl_shader *shader);

diff --git a/src/glsl/lower_variable_index_to_cond_assign.cpp b/src/glsl/lower_variable_index_to_cond_assign.cpp
index 4a6a76c..1ab3afe 100644 (file)
--- a/src/glsl/lower_variable_index_to_cond_assign.cpp
+++ b/src/glsl/lower_variable_index_to_cond_assign.cpp
@@ -335,12 +335,14 @@ struct switch_generator
 
 class variable_index_to_cond_assign_visitor : public ir_rvalue_visitor {
 public:
-   variable_index_to_cond_assign_visitor(bool lower_input,
-                                        bool lower_output,
-                                        bool lower_temp,
-                                        bool lower_uniform)
+   variable_index_to_cond_assign_visitor(gl_shader_stage stage,
+                                         bool lower_input,
+                                         bool lower_output,
+                                         bool lower_temp,
+                                         bool lower_uniform)
    {
       this->progress = false;
+      this->stage = stage;
       this->lower_inputs = lower_input;
       this->lower_outputs = lower_output;
       this->lower_temps = lower_temp;
@@ -348,6 +350,8 @@ public:
    }
 
    bool progress;
+
+   gl_shader_stage stage;
    bool lower_inputs;
    bool lower_outputs;
    bool lower_temps;
@@ -369,18 +373,44 @@ public:
       case ir_var_auto:
       case ir_var_temporary:
         return this->lower_temps;
+
       case ir_var_uniform:
       case ir_var_shader_storage:
         return this->lower_uniforms;
+
       case ir_var_function_in:
       case ir_var_const_in:
          return this->lower_temps;
+
       case ir_var_shader_in:
+         /* The input array size is unknown at compiler time for non-patch
+          * inputs in TCS and TES. The arrays are sized to
+          * the implementation-dependent limit "gl_MaxPatchVertices", but
+          * the real size is stored in the "gl_PatchVerticesIn" built-in
+          * uniform.
+          *
+          * The TCS input array size is specified by
+          * glPatchParameteri(GL_PATCH_VERTICES).
+          *
+          * The TES input array size is specified by the "vertices" output
+          * layout qualifier in TCS.
+          */
+         if ((stage == MESA_SHADER_TESS_CTRL ||
+              stage == MESA_SHADER_TESS_EVAL) && !var->data.patch)
+            return false;
          return this->lower_inputs;
+
       case ir_var_function_out:
+         /* TCS non-patch outputs can only be indexed with "gl_InvocationID".
+          * Other expressions are not allowed.
+          */
+         if (stage == MESA_SHADER_TESS_CTRL && !var->data.patch)
+            return false;
          return this->lower_temps;
+
       case ir_var_shader_out:
          return this->lower_outputs;
+
       case ir_var_function_inout:
         return this->lower_temps;
       }
@@ -523,16 +553,18 @@ public:
 } /* anonymous namespace */
 
 bool
-lower_variable_index_to_cond_assign(exec_list *instructions,
-                                   bool lower_input,
-                                   bool lower_output,
-                                   bool lower_temp,
-                                   bool lower_uniform)
+lower_variable_index_to_cond_assign(gl_shader_stage stage,
+                                    exec_list *instructions,
+                                    bool lower_input,
+                                    bool lower_output,
+                                    bool lower_temp,
+                                    bool lower_uniform)
 {
-   variable_index_to_cond_assign_visitor v(lower_input,
-                                          lower_output,
-                                          lower_temp,
-                                          lower_uniform);
+   variable_index_to_cond_assign_visitor v(stage,
+                                           lower_input,
+                                           lower_output,
+                                           lower_temp,
+                                           lower_uniform);
 
    /* Continue lowering until no progress is made.  If there are multiple
     * levels of indirection (e.g., non-constant indexing of array elements and

diff --git a/src/glsl/test_optpass.cpp b/src/glsl/test_optpass.cpp
index ac3e3f4..fed1fab 100644 (file)
--- a/src/glsl/test_optpass.cpp
+++ b/src/glsl/test_optpass.cpp
@@ -124,7 +124,8 @@ do_optimization(struct exec_list *ir, const char *optimization,
    } else if (sscanf(optimization, "lower_variable_index_to_cond_assign "
                      "( %d , %d , %d , %d ) ", &int_0, &int_1, &int_2,
                      &int_3) == 4) {
-      return lower_variable_index_to_cond_assign(ir, int_0 != 0, int_1 != 0,
+      return lower_variable_index_to_cond_assign(MESA_SHADER_VERTEX, ir,
+                                                 int_0 != 0, int_1 != 0,
                                                  int_2 != 0, int_3 != 0);
    } else if (sscanf(optimization, "lower_quadop_vector ( %d ) ",
                      &int_0) == 1) {

diff --git a/src/mesa/drivers/dri/i965/brw_shader.cpp b/src/mesa/drivers/dri/i965/brw_shader.cpp
index d66baf3..379ff4a 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_shader.cpp
+++ b/src/mesa/drivers/dri/i965/brw_shader.cpp
@@ -233,7 +233,8 @@ brw_lower_packing_builtins(struct brw_context *brw,
 }
 
 static void
-process_glsl_ir(struct brw_context *brw,
+process_glsl_ir(gl_shader_stage stage,
+                struct brw_context *brw,
                 struct gl_shader_program *shader_prog,
                 struct gl_shader *shader)
 {
@@ -281,7 +282,8 @@ process_glsl_ir(struct brw_context *brw,
    lower_quadop_vector(shader->ir, false);
 
    bool lowered_variable_indexing =
-      lower_variable_index_to_cond_assign(shader->ir,
+      lower_variable_index_to_cond_assign((gl_shader_stage)stage,
+                                          shader->ir,
                                           options->EmitNoIndirectInput,
                                           options->EmitNoIndirectOutput,
                                           options->EmitNoIndirectTemp,
@@ -358,7 +360,7 @@ brw_link_shader(struct gl_context *ctx, struct gl_shader_program *shProg)
 
       _mesa_copy_linked_program_data((gl_shader_stage) stage, shProg, prog);
 
-      process_glsl_ir(brw, shProg, shader);
+      process_glsl_ir((gl_shader_stage) stage, brw, shProg, shader);
 
       /* Make a pass over the IR to add state references for any built-in
        * uniforms that are used.  This has to be done now (during linking).

diff --git a/src/mesa/program/ir_to_mesa.cpp b/src/mesa/program/ir_to_mesa.cpp
index 2bd212e..cadbf49 100644 (file)
--- a/src/mesa/program/ir_to_mesa.cpp
+++ b/src/mesa/program/ir_to_mesa.cpp
@@ -2910,7 +2910,7 @@ _mesa_ir_link_shader(struct gl_context *ctx, struct gl_shader_program *prog)
         if (options->EmitNoIndirectInput || options->EmitNoIndirectOutput
             || options->EmitNoIndirectTemp || options->EmitNoIndirectUniform)
           progress =
-            lower_variable_index_to_cond_assign(ir,
+            lower_variable_index_to_cond_assign(prog->_LinkedShaders[i]->Stage, ir,
                                                 options->EmitNoIndirectInput,
                                                 options->EmitNoIndirectOutput,
                                                 options->EmitNoIndirectTemp,

diff --git a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
index c9d40c5..b727c5e 100644 (file)
--- a/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
+++ b/src/mesa/state_tracker/st_glsl_to_tgsi.cpp
@@ -5833,7 +5833,7 @@ st_link_shader(struct gl_context *ctx, struct gl_shader_program *prog)
        */
       if (options->EmitNoIndirectInput || options->EmitNoIndirectOutput ||
           options->EmitNoIndirectTemp || options->EmitNoIndirectUniform) {
-         lower_variable_index_to_cond_assign(ir,
+         lower_variable_index_to_cond_assign(prog->_LinkedShaders[i]->Stage, ir,
                                              options->EmitNoIndirectInput,
                                              options->EmitNoIndirectOutput,
                                              options->EmitNoIndirectTemp,