i965/fs: Add a shader_stats struct.

It'll grow further, and we'd like to avoid adding an additional
parameter to fs_generator() for each new piece of data.

v2 (idr): Rebase on 17 months.  Track a visitor instead of a cfg.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>

src/intel/compiler/brw_fs.cpp

@@ -8108,7 +8108,7 @@ brw_compile_fs(const struct brw_compiler *compiler, void *log_data,
    }
 
    fs_generator g(compiler, log_data, mem_ctx, &prog_data->base,
-                  v8.promoted_constants, v8.runtime_check_aads_emit,
+                  v8.shader_stats, v8.runtime_check_aads_emit,
                   MESA_SHADER_FRAGMENT);
 
    if (unlikely(INTEL_DEBUG & DEBUG_WM)) {
@@ -8258,9 +8258,8 @@ brw_compile_cs(const struct brw_compiler *compiler, void *log_data,
    unsigned max_dispatch_width = 32;
 
    fs_visitor *v8 = NULL, *v16 = NULL, *v32 = NULL;
-   cfg_t *cfg = NULL;
+   fs_visitor *v = NULL;
    const char *fail_msg = NULL;
-   unsigned promoted_constants = 0;
 
    if ((int)key->base.subgroup_size_type >= (int)BRW_SUBGROUP_SIZE_REQUIRE_8) {
       /* These enum values are expressly chosen to be equal to the subgroup
@@ -8294,10 +8293,9 @@ brw_compile_cs(const struct brw_compiler *compiler, void *log_data,
          /* We should always be able to do SIMD32 for compute shaders */
          assert(v8->max_dispatch_width >= 32);
 
-         cfg = v8->cfg;
+         v = v8;
          cs_set_simd_size(prog_data, 8);
          cs_fill_push_const_info(compiler->devinfo, prog_data);
-         promoted_constants = v8->promoted_constants;
       }
    }
 
@@ -8317,7 +8315,7 @@ brw_compile_cs(const struct brw_compiler *compiler, void *log_data,
          compiler->shader_perf_log(log_data,
                                    "SIMD16 shader failed to compile: %s",
                                    v16->fail_msg);
-         if (!cfg) {
+         if (!v) {
             fail_msg =
                "Couldn't generate SIMD16 program and not "
                "enough threads for SIMD8";
@@ -8326,10 +8324,9 @@ brw_compile_cs(const struct brw_compiler *compiler, void *log_data,
          /* We should always be able to do SIMD32 for compute shaders */
          assert(v16->max_dispatch_width >= 32);
 
-         cfg = v16->cfg;
+         v = v16;
          cs_set_simd_size(prog_data, 16);
          cs_fill_push_const_info(compiler->devinfo, prog_data);
-         promoted_constants = v16->promoted_constants;
       }
    }
 
@@ -8354,27 +8351,27 @@ brw_compile_cs(const struct brw_compiler *compiler, void *log_data,
          compiler->shader_perf_log(log_data,
                                    "SIMD32 shader failed to compile: %s",
                                    v16->fail_msg);
-         if (!cfg) {
+         if (!v) {
             fail_msg =
                "Couldn't generate SIMD32 program and not "
                "enough threads for SIMD16";
          }
       } else {
-         cfg = v32->cfg;
+         v = v32;
          cs_set_simd_size(prog_data, 32);
          cs_fill_push_const_info(compiler->devinfo, prog_data);
-         promoted_constants = v32->promoted_constants;
       }
    }
 
    const unsigned *ret = NULL;
-   if (unlikely(cfg == NULL)) {
+   if (unlikely(v == NULL)) {
       assert(fail_msg);
       if (error_str)
          *error_str = ralloc_strdup(mem_ctx, fail_msg);
    } else {
       fs_generator g(compiler, log_data, mem_ctx, &prog_data->base,
-                     promoted_constants, false, MESA_SHADER_COMPUTE);
+                     v->shader_stats, v->runtime_check_aads_emit,
+                     MESA_SHADER_COMPUTE);
       if (INTEL_DEBUG & DEBUG_CS) {
          char *name = ralloc_asprintf(mem_ctx, "%s compute shader %s",
                                       src_shader->info.label ?
@@ -8383,7 +8380,7 @@ brw_compile_cs(const struct brw_compiler *compiler, void *log_data,
          g.enable_debug(name);
       }
 
-      g.generate_code(cfg, prog_data->simd_size);
+      g.generate_code(v->cfg, prog_data->simd_size);
 
       ret = g.get_assembly();
    }

src/intel/compiler/brw_fs.h

@@ -52,6 +52,10 @@ offset(const fs_reg &reg, const brw::fs_builder &bld, unsigned delta)
 
 #define UBO_START ((1 << 16) - 4)
 
+struct shader_stats {
+   unsigned promoted_constants;
+};
+
 /**
  * The fragment shader front-end.
  *
@@ -389,7 +393,8 @@ public:
 
    int shader_time_index;
 
-   unsigned promoted_constants;
+   struct shader_stats shader_stats;
+
    brw::fs_builder bld;
 
 private:
@@ -413,7 +418,7 @@ public:
    fs_generator(const struct brw_compiler *compiler, void *log_data,
                 void *mem_ctx,
                 struct brw_stage_prog_data *prog_data,
-                unsigned promoted_constants,
+                struct shader_stats shader_stats,
                 bool runtime_check_aads_emit,
                 gl_shader_stage stage);
    ~fs_generator();
@@ -517,7 +522,7 @@ private:
    unsigned dispatch_width; /**< 8, 16 or 32 */
 
    exec_list discard_halt_patches;
-   unsigned promoted_constants;
+   struct shader_stats shader_stats;
    bool runtime_check_aads_emit;
    bool debug_flag;
    const char *shader_name;

src/intel/compiler/brw_fs_combine_constants.cpp

@@ -442,7 +442,7 @@ fs_visitor::opt_combine_constants()
 
       reg.offset += imm->size * width;
    }
-   promoted_constants = table.len;
+   shader_stats.promoted_constants = table.len;
 
    /* Rewrite the immediate sources to refer to the new GRFs. */
    for (int i = 0; i < table.len; i++) {

src/intel/compiler/brw_fs_generator.cpp

@@ -183,14 +183,14 @@ brw_reg_from_fs_reg(const struct gen_device_info *devinfo, fs_inst *inst,
 fs_generator::fs_generator(const struct brw_compiler *compiler, void *log_data,
                            void *mem_ctx,
                            struct brw_stage_prog_data *prog_data,
-                           unsigned promoted_constants,
+                           struct shader_stats shader_stats,
                            bool runtime_check_aads_emit,
                            gl_shader_stage stage)
 
    : compiler(compiler), log_data(log_data),
      devinfo(compiler->devinfo),
      prog_data(prog_data),
-     promoted_constants(promoted_constants),
+     shader_stats(shader_stats),
      runtime_check_aads_emit(runtime_check_aads_emit), debug_flag(false),
      stage(stage), mem_ctx(mem_ctx)
 {
@@ -2253,7 +2253,7 @@ fs_generator::generate_code(const cfg_t *cfg, int dispatch_width)
               "SIMD%d shader: %d instructions. %d loops. %u cycles. %d:%d spills:fills. Promoted %u constants. Compacted %d to %d"
               " bytes (%.0f%%)\n",
               shader_name, dispatch_width, before_size / 16, loop_count, cfg->cycle_count,
-              spill_count, fill_count, promoted_constants, before_size, after_size,
+              spill_count, fill_count, shader_stats.promoted_constants, before_size, after_size,
               100.0f * (before_size - after_size) / before_size);
 
       dump_assembly(p->store, disasm_info);
@@ -2268,7 +2268,7 @@ fs_generator::generate_code(const cfg_t *cfg, int dispatch_width)
                               _mesa_shader_stage_to_abbrev(stage),
                               dispatch_width, before_size / 16,
                               loop_count, cfg->cycle_count, spill_count,
-                              fill_count, promoted_constants, before_size,
+                              fill_count, shader_stats.promoted_constants, before_size,
                               after_size);
 
    return start_offset;

src/intel/compiler/brw_fs_visitor.cpp

@@ -952,7 +952,7 @@ fs_visitor::init()
    this->pull_constant_loc = NULL;
    this->push_constant_loc = NULL;
 
-   this->promoted_constants = 0,
+   this->shader_stats.promoted_constants = 0,
 
    this->grf_used = 0;
    this->spilled_any_registers = false;

src/intel/compiler/brw_shader.cpp

@@ -1335,7 +1335,7 @@ brw_compile_tes(const struct brw_compiler *compiler,
       prog_data->base.dispatch_mode = DISPATCH_MODE_SIMD8;
 
       fs_generator g(compiler, log_data, mem_ctx,
-                     &prog_data->base.base, v.promoted_constants, false,
+                     &prog_data->base.base, v.shader_stats, false,
                      MESA_SHADER_TESS_EVAL);
       if (unlikely(INTEL_DEBUG & DEBUG_TES)) {
          g.enable_debug(ralloc_asprintf(mem_ctx,

src/intel/compiler/brw_vec4.cpp

@@ -2975,7 +2975,7 @@ brw_compile_vs(const struct brw_compiler *compiler, void *log_data,
       prog_data->base.base.dispatch_grf_start_reg = v.payload.num_regs;
 
       fs_generator g(compiler, log_data, mem_ctx,
-                     &prog_data->base.base, v.promoted_constants,
+                     &prog_data->base.base, v.shader_stats,
                      v.runtime_check_aads_emit, MESA_SHADER_VERTEX);
       if (INTEL_DEBUG & DEBUG_VS) {
          const char *debug_name =

src/intel/compiler/brw_vec4_gs_visitor.cpp

@@ -856,7 +856,7 @@ brw_compile_gs(const struct brw_compiler *compiler, void *log_data,
          prog_data->base.base.dispatch_grf_start_reg = v.payload.num_regs;
 
          fs_generator g(compiler, log_data, mem_ctx,
-                        &prog_data->base.base, v.promoted_constants,
+                        &prog_data->base.base, v.shader_stats,
                         false, MESA_SHADER_GEOMETRY);
          if (unlikely(INTEL_DEBUG & DEBUG_GS)) {
             const char *label =

src/intel/compiler/brw_vec4_tcs.cpp

@@ -487,7 +487,7 @@ brw_compile_tcs(const struct brw_compiler *compiler,
       prog_data->base.base.dispatch_grf_start_reg = v.payload.num_regs;
 
       fs_generator g(compiler, log_data, mem_ctx,
-                     &prog_data->base.base, v.promoted_constants, false,
+                     &prog_data->base.base, v.shader_stats, false,
                      MESA_SHADER_TESS_CTRL);
       if (unlikely(INTEL_DEBUG & DEBUG_TCS)) {
          g.enable_debug(ralloc_asprintf(mem_ctx,