i965: Always use push constants for pixel shaderfoo

Works around a HW bug resulting into GPU hang on Broxton.

Signed-off-by: Topi Pohjolainen <topi.pohjolainen@intel.com>

7 files changed, 24 insertions, 6 deletions

diff --git a/src/intel/blorp/blorp.c b/src/intel/blorp/blorp.c
index a426a030d2..b2ca71b0fe 100644
--- a/src/intel/blorp/blorp.c
+++ b/src/intel/blorp/blorp.c
@@ -194,7 +194,7 @@ blorp_compile_fs(struct blorp_context *blorp, void *mem_ctx,
    const unsigned *program =
       brw_compile_fs(compiler, blorp->driver_ctx, mem_ctx, wm_key,
                      wm_prog_data, nir, NULL, -1, -1, false, use_repclear,
-                     NULL, program_size, NULL);
+                     false, NULL, program_size, NULL);
 
    return program;
 }
diff --git a/src/intel/compiler/brw_compiler.h b/src/intel/compiler/brw_compiler.h
index 6753a8daf0..74efe1cd37 100644
--- a/src/intel/compiler/brw_compiler.h
+++ b/src/intel/compiler/brw_compiler.h
@@ -1122,7 +1122,8 @@ brw_compile_fs(const struct brw_compiler *compiler, void *log_data,
                int shader_time_index8,
                int shader_time_index16,
                bool allow_spilling,
-               bool use_rep_send, struct brw_vue_map *vue_map,
+               bool use_rep_send, bool force_enable_push_constants,
+               struct brw_vue_map *vue_map,
                unsigned *final_assembly_size,
                char **error_str);
 
diff --git a/src/intel/compiler/brw_fs.cpp b/src/intel/compiler/brw_fs.cpp
index eb9b4c3890..d3d2d3dd42 100644
--- a/src/intel/compiler/brw_fs.cpp
+++ b/src/intel/compiler/brw_fs.cpp
@@ -2082,6 +2082,12 @@ fs_visitor::assign_constant_locations()
    if (thread_local_id_index >= 0)
       push_constant_loc[thread_local_id_index] = num_push_constants++;
 
+   if (stage == MESA_SHADER_FRAGMENT && num_push_constants == 0 &&
+       force_enable_push_constants) {
+      assert(uniforms == 0);
+      num_push_constants = 1;
+   }
+
    /* As the uniforms are going to be reordered, take the data from a temporary
     * copy of the original param[].
     */
@@ -6505,7 +6511,8 @@ brw_compile_fs(const struct brw_compiler *compiler, void *log_data,
                struct gl_program *prog,
                int shader_time_index8, int shader_time_index16,
                bool allow_spilling,
-               bool use_rep_send, struct brw_vue_map *vue_map,
+               bool use_rep_send, bool force_enable_push_constants,
+               struct brw_vue_map *vue_map,
                unsigned *final_assembly_size,
                char **error_str)
 {
@@ -6560,6 +6567,7 @@ brw_compile_fs(const struct brw_compiler *compiler, void *log_data,
    fs_visitor v8(compiler, log_data, mem_ctx, key,
                  &prog_data->base, prog, shader, 8,
                  shader_time_index8);
+   v8.force_enable_push_constants = force_enable_push_constants;
    if (!v8.run_fs(allow_spilling, false /* do_rep_send */)) {
       if (error_str)
          *error_str = ralloc_strdup(mem_ctx, v8.fail_msg);
@@ -6577,6 +6585,7 @@ brw_compile_fs(const struct brw_compiler *compiler, void *log_data,
       fs_visitor v16(compiler, log_data, mem_ctx, key,
                      &prog_data->base, prog, shader, 16,
                      shader_time_index16);
+      v16.force_enable_push_constants = force_enable_push_constants;
       v16.import_uniforms(&v8);
       if (!v16.run_fs(allow_spilling, use_rep_send)) {
          compiler->shader_perf_log(log_data,
diff --git a/src/intel/compiler/brw_fs.h b/src/intel/compiler/brw_fs.h
index f1ba193de7..0736cb35d2 100644
--- a/src/intel/compiler/brw_fs.h
+++ b/src/intel/compiler/brw_fs.h
@@ -348,6 +348,7 @@ public:
 
    bool source_depth_to_render_target;
    bool runtime_check_aads_emit;
+   bool force_enable_push_constants;
 
    fs_reg pixel_x;
    fs_reg pixel_y;
diff --git a/src/intel/vulkan/anv_pipeline.c b/src/intel/vulkan/anv_pipeline.c
index 94e99d8437..4dc5cc4fc7 100644
--- a/src/intel/vulkan/anv_pipeline.c
+++ b/src/intel/vulkan/anv_pipeline.c
@@ -928,7 +928,8 @@ anv_pipeline_compile_fs(struct anv_pipeline *pipeline,
       unsigned code_size;
       const unsigned *shader_code =
          brw_compile_fs(compiler, NULL, mem_ctx, &key, &prog_data, nir,
-                        NULL, -1, -1, true, false, NULL, &code_size, NULL);
+                        NULL, -1, -1, true, false, false,
+                        NULL, &code_size, NULL);
       if (shader_code == NULL) {
          ralloc_free(mem_ctx);
          return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
diff --git a/src/mesa/drivers/dri/i965/brw_wm.c b/src/mesa/drivers/dri/i965/brw_wm.c
index 92354e464c..d03f93b325 100644
--- a/src/mesa/drivers/dri/i965/brw_wm.c
+++ b/src/mesa/drivers/dri/i965/brw_wm.c
@@ -193,7 +193,7 @@ brw_codegen_wm_prog(struct brw_context *brw,
    program = brw_compile_fs(brw->screen->compiler, brw, mem_ctx,
                             key, &prog_data, fp->program.nir,
                             &fp->program, st_index8, st_index16,
-                            true, false, vue_map,
+                            true, false, true, vue_map,
                             &program_size, &error_str);
 
    if (program == NULL) {
diff --git a/src/mesa/drivers/dri/i965/gen6_constant_state.c b/src/mesa/drivers/dri/i965/gen6_constant_state.c
index 72f00d5640..4f20aaa45e 100644
--- a/src/mesa/drivers/dri/i965/gen6_constant_state.c
+++ b/src/mesa/drivers/dri/i965/gen6_constant_state.c
@@ -81,9 +81,15 @@ gen6_upload_push_constants(struct brw_context *brw,
        * Also _NEW_TRANSFORM -- we may reference clip planes other than as a
        * side effect of dereferencing uniforms, so _NEW_PROGRAM_CONSTANTS
        * wouldn't be set for them.
+       *
+       * In order to workaround a GPU hang on Broxton pixel shader gets
+       * always programmed with push constant support. In such case nr_params
+       * is non-zero but the backing storage doesn't exist - simply skip the
+       * copy and use unitialised which doesn't really get used by the shader.
        */
       for (i = 0; i < prog_data->nr_params; i++) {
-         param[i] = *prog_data->param[i];
+          if (prog_data->param[i])
+             param[i] = *prog_data->param[i];
       }
 
       if (0) {