intel/ir: Use brw::performance object instead of CFG cycle counts for codegen stats.

These should be more accurate than the current cycle counts, since
among other things they consider the effect of post-scheduling passes
like the software scoreboard on TGL.  In addition it will enable us to
clean up some of the now redundant cycle-count estimation
functionality in the instruction scheduler.

Reviewed-by: Kenneth Graunke <[email protected]>

9 files changed, 45 insertions, 22 deletions

diff --git a/src/intel/compiler/brw_fs.cpp b/src/intel/compiler/brw_fs.cpp
index 57d12139233..ccefdb081fd 100644
--- a/src/intel/compiler/brw_fs.cpp
+++ b/src/intel/compiler/brw_fs.cpp
@@ -8845,21 +8845,24 @@ brw_compile_fs(const struct brw_compiler *compiler, void *log_data,
 
    if (simd8_cfg) {
       prog_data->dispatch_8 = true;
-      g.generate_code(simd8_cfg, 8, v8->shader_stats, stats);
+      g.generate_code(simd8_cfg, 8, v8->shader_stats,
+                      v8->performance_analysis.require(), stats);
       stats = stats ? stats + 1 : NULL;
    }
 
    if (simd16_cfg) {
       prog_data->dispatch_16 = true;
-      prog_data->prog_offset_16 = g.generate_code(simd16_cfg, 16,
-                                                  v16->shader_stats, stats);
+      prog_data->prog_offset_16 = g.generate_code(
+         simd16_cfg, 16, v16->shader_stats,
+         v16->performance_analysis.require(), stats);
       stats = stats ? stats + 1 : NULL;
    }
 
    if (simd32_cfg) {
       prog_data->dispatch_32 = true;
-      prog_data->prog_offset_32 = g.generate_code(simd32_cfg, 32,
-                                                  v32->shader_stats, stats);
+      prog_data->prog_offset_32 = g.generate_code(
+         simd32_cfg, 32, v32->shader_stats,
+         v32->performance_analysis.require(), stats);
       stats = stats ? stats + 1 : NULL;
    }
 
@@ -9118,7 +9121,8 @@ brw_compile_cs(const struct brw_compiler *compiler, void *log_data,
          g.enable_debug(name);
       }
 
-      g.generate_code(v->cfg, prog_data->simd_size, v->shader_stats, stats);
+      g.generate_code(v->cfg, prog_data->simd_size, v->shader_stats,
+                      v->performance_analysis.require(), stats);
 
       ret = g.get_assembly();
    }
diff --git a/src/intel/compiler/brw_fs.h b/src/intel/compiler/brw_fs.h
index 4e8f8ccac78..b9783eb2285 100644
--- a/src/intel/compiler/brw_fs.h
+++ b/src/intel/compiler/brw_fs.h
@@ -477,6 +477,7 @@ public:
    void enable_debug(const char *shader_name);
    int generate_code(const cfg_t *cfg, int dispatch_width,
                      struct shader_stats shader_stats,
+                     const brw::performance &perf,
                      struct brw_compile_stats *stats);
    const unsigned *get_assembly();
 
diff --git a/src/intel/compiler/brw_fs_generator.cpp b/src/intel/compiler/brw_fs_generator.cpp
index d7464c95297..dc524a61e1d 100644
--- a/src/intel/compiler/brw_fs_generator.cpp
+++ b/src/intel/compiler/brw_fs_generator.cpp
@@ -1715,6 +1715,7 @@ fs_generator::enable_debug(const char *shader_name)
 int
 fs_generator::generate_code(const cfg_t *cfg, int dispatch_width,
                             struct shader_stats shader_stats,
+                            const brw::performance &perf,
                             struct brw_compile_stats *stats)
 {
    /* align to 64 byte boundary. */
@@ -2462,7 +2463,7 @@ fs_generator::generate_code(const cfg_t *cfg, int dispatch_width,
               "Compacted %d to %d bytes (%.0f%%)\n",
               shader_name, sha1buf,
               dispatch_width, before_size / 16,
-              loop_count, cfg->cycle_count,
+              loop_count, perf.latency,
               spill_count, fill_count, send_count,
               shader_stats.scheduler_mode,
               shader_stats.promoted_constants,
@@ -2487,7 +2488,7 @@ fs_generator::generate_code(const cfg_t *cfg, int dispatch_width,
                               "compacted %d to %d bytes.",
                               _mesa_shader_stage_to_abbrev(stage),
                               dispatch_width, before_size / 16 - nop_count,
-                              loop_count, cfg->cycle_count,
+                              loop_count, perf.latency,
                               spill_count, fill_count, send_count,
                               shader_stats.scheduler_mode,
                               shader_stats.promoted_constants,
@@ -2497,7 +2498,7 @@ fs_generator::generate_code(const cfg_t *cfg, int dispatch_width,
       stats->instructions = before_size / 16 - nop_count;
       stats->sends = send_count;
       stats->loops = loop_count;
-      stats->cycles = cfg->cycle_count;
+      stats->cycles = perf.latency;
       stats->spills = spill_count;
       stats->fills = fill_count;
    }
diff --git a/src/intel/compiler/brw_shader.cpp b/src/intel/compiler/brw_shader.cpp
index b0a75f067a6..470d77227ab 100644
--- a/src/intel/compiler/brw_shader.cpp
+++ b/src/intel/compiler/brw_shader.cpp
@@ -1371,7 +1371,8 @@ brw_compile_tes(const struct brw_compiler *compiler,
                                         nir->info.name));
       }
 
-      g.generate_code(v.cfg, 8, v.shader_stats, stats);
+      g.generate_code(v.cfg, 8, v.shader_stats,
+                      v.performance_analysis.require(), stats);
 
       assembly = g.get_assembly();
    } else {
@@ -1387,7 +1388,9 @@ brw_compile_tes(const struct brw_compiler *compiler,
 	 v.dump_instructions();
 
       assembly = brw_vec4_generate_assembly(compiler, log_data, mem_ctx, nir,
-                                            &prog_data->base, v.cfg, stats);
+                                            &prog_data->base, v.cfg,
+                                            v.performance_analysis.require(),
+                                            stats);
    }
 
    return assembly;
diff --git a/src/intel/compiler/brw_vec4.cpp b/src/intel/compiler/brw_vec4.cpp
index 6f05046bca6..864ece7e636 100644
--- a/src/intel/compiler/brw_vec4.cpp
+++ b/src/intel/compiler/brw_vec4.cpp
@@ -2999,7 +2999,8 @@ brw_compile_vs(const struct brw_compiler *compiler, void *log_data,
 
          g.enable_debug(debug_name);
       }
-      g.generate_code(v.cfg, 8, v.shader_stats, stats);
+      g.generate_code(v.cfg, 8, v.shader_stats,
+                      v.performance_analysis.require(), stats);
       assembly = g.get_assembly();
    }
 
@@ -3017,7 +3018,9 @@ brw_compile_vs(const struct brw_compiler *compiler, void *log_data,
 
       assembly = brw_vec4_generate_assembly(compiler, log_data, mem_ctx,
                                             shader, &prog_data->base,
-                                            v.cfg, stats);
+                                            v.cfg,
+                                            v.performance_analysis.require(),
+                                            stats);
    }
 
    return assembly;
diff --git a/src/intel/compiler/brw_vec4.h b/src/intel/compiler/brw_vec4.h
index aa93b05d5af..73c18b6748c 100644
--- a/src/intel/compiler/brw_vec4.h
+++ b/src/intel/compiler/brw_vec4.h
@@ -48,6 +48,7 @@ brw_vec4_generate_assembly(const struct brw_compiler *compiler,
                            const nir_shader *nir,
                            struct brw_vue_prog_data *prog_data,
                            const struct cfg_t *cfg,
+                           const brw::performance &perf,
                            struct brw_compile_stats *stats);
 
 #ifdef __cplusplus
diff --git a/src/intel/compiler/brw_vec4_generator.cpp b/src/intel/compiler/brw_vec4_generator.cpp
index bdc45371c4e..e95a70d8605 100644
--- a/src/intel/compiler/brw_vec4_generator.cpp
+++ b/src/intel/compiler/brw_vec4_generator.cpp
@@ -1498,6 +1498,7 @@ generate_code(struct brw_codegen *p,
               const nir_shader *nir,
               struct brw_vue_prog_data *prog_data,
               const struct cfg_t *cfg,
+              const performance &perf,
               struct brw_compile_stats *stats)
 {
    const struct gen_device_info *devinfo = p->devinfo;
@@ -2220,7 +2221,7 @@ generate_code(struct brw_codegen *p,
 
       fprintf(stderr, "%s vec4 shader: %d instructions. %d loops. %u cycles. %d:%d "
                      "spills:fills, %u sends. Compacted %d to %d bytes (%.0f%%)\n",
-            stage_abbrev, before_size / 16, loop_count, cfg->cycle_count,
+            stage_abbrev, before_size / 16, loop_count, perf.latency,
             spill_count, fill_count, send_count, before_size, after_size,
             100.0f * (before_size - after_size) / before_size);
 
@@ -2239,14 +2240,14 @@ generate_code(struct brw_codegen *p,
                               "%d:%d spills:fills, %u sends, "
                               "compacted %d to %d bytes.",
                               stage_abbrev, before_size / 16,
-                              loop_count, cfg->cycle_count, spill_count,
+                              loop_count, perf.latency, spill_count,
                               fill_count, send_count, before_size, after_size);
    if (stats) {
       stats->dispatch_width = 0;
       stats->instructions = before_size / 16;
       stats->sends = send_count;
       stats->loops = loop_count;
-      stats->cycles = cfg->cycle_count;
+      stats->cycles = perf.latency;
       stats->spills = spill_count;
       stats->fills = fill_count;
    }
@@ -2259,13 +2260,14 @@ brw_vec4_generate_assembly(const struct brw_compiler *compiler,
                            const nir_shader *nir,
                            struct brw_vue_prog_data *prog_data,
                            const struct cfg_t *cfg,
+                           const performance &perf,
                            struct brw_compile_stats *stats)
 {
    struct brw_codegen *p = rzalloc(mem_ctx, struct brw_codegen);
    brw_init_codegen(compiler->devinfo, p, mem_ctx);
    brw_set_default_access_mode(p, BRW_ALIGN_16);
 
-   generate_code(p, compiler, log_data, nir, prog_data, cfg, stats);
+   generate_code(p, compiler, log_data, nir, prog_data, cfg, perf, stats);
 
    return brw_get_program(p, &prog_data->base.program_size);
 }
diff --git a/src/intel/compiler/brw_vec4_gs_visitor.cpp b/src/intel/compiler/brw_vec4_gs_visitor.cpp
index a0b78eb4d48..ce341806c7b 100644
--- a/src/intel/compiler/brw_vec4_gs_visitor.cpp
+++ b/src/intel/compiler/brw_vec4_gs_visitor.cpp
@@ -865,7 +865,8 @@ brw_compile_gs(const struct brw_compiler *compiler, void *log_data,
                                          label, shader->info.name);
             g.enable_debug(name);
          }
-         g.generate_code(v.cfg, 8, v.shader_stats, stats);
+         g.generate_code(v.cfg, 8, v.shader_stats,
+                         v.performance_analysis.require(), stats);
          return g.get_assembly();
       }
    }
@@ -897,7 +898,9 @@ brw_compile_gs(const struct brw_compiler *compiler, void *log_data,
             ralloc_free(param);
             return brw_vec4_generate_assembly(compiler, log_data, mem_ctx,
                                               shader, &prog_data->base,
-                                              v.cfg, stats);
+                                              v.cfg,
+                                              v.performance_analysis.require(),
+                                              stats);
          } else {
             /* These variables could be modified by the execution of the GS
              * visitor if it packed the uniforms in the push constant buffer.
@@ -960,7 +963,9 @@ brw_compile_gs(const struct brw_compiler *compiler, void *log_data,
          *error_str = ralloc_strdup(mem_ctx, gs->fail_msg);
    } else {
       ret = brw_vec4_generate_assembly(compiler, log_data, mem_ctx, shader,
-                                       &prog_data->base, gs->cfg, stats);
+                                       &prog_data->base, gs->cfg,
+                                       gs->performance_analysis.require(),
+                                       stats);
    }
 
    delete gs;
diff --git a/src/intel/compiler/brw_vec4_tcs.cpp b/src/intel/compiler/brw_vec4_tcs.cpp
index 26dc4f18040..852c5a9865c 100644
--- a/src/intel/compiler/brw_vec4_tcs.cpp
+++ b/src/intel/compiler/brw_vec4_tcs.cpp
@@ -480,7 +480,8 @@ brw_compile_tcs(const struct brw_compiler *compiler,
                                         nir->info.name));
       }
 
-      g.generate_code(v.cfg, 8, v.shader_stats, stats);
+      g.generate_code(v.cfg, 8, v.shader_stats,
+                      v.performance_analysis.require(), stats);
 
       assembly = g.get_assembly();
    } else {
@@ -497,7 +498,9 @@ brw_compile_tcs(const struct brw_compiler *compiler,
 
 
       assembly = brw_vec4_generate_assembly(compiler, log_data, mem_ctx, nir,
-                                            &prog_data->base, v.cfg, stats);
+                                            &prog_data->base, v.cfg,
+                                            v.performance_analysis.require(),
+                                            stats);
    }
 
    return assembly;