brw: Use debug archive file with INTEL_DEBUG=mda

Instead of dumping multiple files with the optimizer passes, write a single
archive file with all the contents.  The actual file is created
by the drivers, so later commits will actually enable the feature in
anv and iris.

This removes the use of INTEL_DEBUG=optimizer (and the corresponding
enum value) in brw.  That environment variable is still used by ELK --
which currently doesn't support mda.

Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29146>

docs/envvars.rst

@@ -565,6 +565,9 @@ Intel driver environment variables
       print instruction hex dump with the disassembly
    ``l3``
       emit messages about the new L3 state during transitions
+   ``mda``
+      generate mda.tar files containing shader at each optimization
+      pass and iteration that make progress (Gfx >= 9)
    ``mesh``
       dump shader assembly for mesh shaders
    ``no8``
@@ -588,7 +591,7 @@ Intel driver environment variables
       disable lossless color compression
    ``optimizer``
       dump shader assembly to files at each optimization pass and
-      iteration that make progress
+      iteration that make progress (Gfx < 9)
    ``pc``
       emit messages about PIPE_CONTROL instruction usage
    ``perf``

src/intel/compiler/brw_compile_bs.cpp

@@ -97,6 +97,7 @@ compile_single_bs(const struct brw_compiler *compiler,
       .needs_register_pressure = stats != NULL,
       .log_data                = params->base.log_data,
       .debug_enabled           = debug_enabled,
+      .archiver                = params->base.archiver,
    };
    brw_shader s(&shader_params);
 

src/intel/compiler/brw_compile_cs.cpp

@@ -221,6 +221,7 @@ brw_compile_cs(const struct brw_compiler *compiler,
          .needs_register_pressure = params->base.stats != NULL,
          .log_data                = params->base.log_data,
          .debug_enabled           = debug_enabled,
+         .archiver                = params->base.archiver,
       };
       v[simd] = std::make_unique<brw_shader>(&shader_params);
       brw_adjust_uniforms(*v[simd]);

src/intel/compiler/brw_compile_fs.cpp

@@ -1595,6 +1595,7 @@ brw_compile_fs(const struct brw_compiler *compiler,
       .needs_register_pressure = params->base.stats != NULL,
       .log_data                = params->base.log_data,
       .debug_enabled           = debug_enabled,
+      .archiver                = params->base.archiver,
    };
 
    if (devinfo->ver < 20) {

src/intel/compiler/brw_compile_gs.cpp

@@ -370,6 +370,7 @@ brw_compile_gs(const struct brw_compiler *compiler,
       .needs_register_pressure = params->base.stats != NULL,
       .log_data                = params->base.log_data,
       .debug_enabled           = debug_enabled,
+      .archiver                = params->base.archiver,
    };
    brw_shader v(&shader_params);
    v.gs.control_data_bits_per_vertex = control_data_bits_per_vertex;

src/intel/compiler/brw_compile_mesh.cpp

@@ -416,6 +416,7 @@ brw_compile_task(const struct brw_compiler *compiler,
          .needs_register_pressure = params->base.stats != NULL,
          .log_data                = params->base.log_data,
          .debug_enabled           = debug_enabled,
+         .archiver                = params->base.archiver,
       };
       v[simd] = std::make_unique<brw_shader>(&shader_params);
 
@@ -1270,6 +1271,7 @@ brw_compile_mesh(const struct brw_compiler *compiler,
          .needs_register_pressure = params->base.stats != NULL,
          .log_data                = params->base.log_data,
          .debug_enabled           = debug_enabled,
+         .archiver                = params->base.archiver,
       };
       v[simd] = std::make_unique<brw_shader>(&shader_params);
 

src/intel/compiler/brw_compile_tcs.cpp

@@ -281,6 +281,7 @@ brw_compile_tcs(const struct brw_compiler *compiler,
       .needs_register_pressure = params->base.stats != NULL,
       .log_data                = params->base.log_data,
       .debug_enabled           = debug_enabled,
+      .archiver                = params->base.archiver,
    };
    brw_shader v(&shader_params);
    if (!run_tcs(v)) {

src/intel/compiler/brw_compile_tes.cpp

@@ -183,6 +183,7 @@ brw_compile_tes(const struct brw_compiler *compiler,
       .needs_register_pressure = params->base.stats != NULL,
       .log_data                = params->base.log_data,
       .debug_enabled           = debug_enabled,
+      .archiver                = params->base.archiver,
    };
    brw_shader v(&shader_params);
    if (!run_tes(v)) {

src/intel/compiler/brw_compile_vs.cpp

@@ -363,6 +363,7 @@ brw_compile_vs(const struct brw_compiler *compiler,
       .needs_register_pressure = params->base.stats != NULL,
       .log_data                = params->base.log_data,
       .debug_enabled           = debug_enabled,
+      .archiver                = params->base.archiver,
    };
    brw_shader v(&shader_params);
    if (!run_vs(v)) {

src/intel/compiler/brw_compiler.h

@@ -27,6 +27,7 @@
 #include "c11/threads.h"
 #include "dev/intel_device_info.h"
 #include "isl/isl.h"
+#include "mda/debug_archiver.h"
 #include "util/macros.h"
 #include "util/mesa-sha1.h"
 #include "util/enum_operators.h"
@@ -1478,6 +1479,8 @@ struct brw_compile_params {
    uint64_t debug_flag;
 
    uint32_t source_hash;
+
+   debug_archiver *archiver;
 };
 
 /**

src/intel/compiler/brw_shader.cpp

@@ -408,7 +408,8 @@ brw_shader::brw_shader(const brw_shader_params *params)
      needs_register_pressure(params->needs_register_pressure),
      dispatch_width(params->dispatch_width),
      max_polygons(params->num_polygons),
-     api_subgroup_size(brw_nir_api_subgroup_size(params->nir, dispatch_width))
+     api_subgroup_size(brw_nir_api_subgroup_size(params->nir, dispatch_width)),
+     archiver(params->archiver)
 {
    assert(api_subgroup_size == 0 ||
          api_subgroup_size == 8 ||
@@ -1009,31 +1010,18 @@ brw_shader::debug_optimizer(const nir_shader *nir,
                             const char *pass_name,
                             int iteration, int pass_num) const
 {
-   /* source_hash is not readily accessible in this context */
-   if (!brw_should_print_shader(nir, DEBUG_OPTIMIZER, 0))
+   if (!archiver)
       return;
 
-   char *filename;
-   int ret = asprintf(&filename, "%s/%s%d-%s-%02d-%02d-%s",
-                      debug_get_option("INTEL_SHADER_OPTIMIZER_PATH", "./"),
-                      _mesa_shader_stage_to_abbrev(stage), dispatch_width, nir->info.name,
+   /* TODO: Add replacement for INTEL_SHADER_OPTIMIZER_PATH. */
+   const char *filename =
+      ralloc_asprintf(mem_ctx, "BRW%d/%02d-%02d-%s",
+                      dispatch_width,
                       iteration, pass_num, pass_name);
-   if (ret == -1)
-      return;
 
-   FILE *file = stderr;
-   if (__normal_user()) {
-      file = fopen(filename, "w");
-      if (!file)
-         file = stderr;
-   }
-
-   brw_print_instructions(*this, file);
-
-   if (file != stderr)
-      fclose(file);
-
-   free(filename);
+   FILE *f = debug_archiver_start_file(archiver, filename);
+   brw_print_instructions(*this, f);
+   debug_archiver_finish_file(archiver);
 }
 
 static uint32_t

src/intel/compiler/brw_shader.h

@@ -77,6 +77,8 @@ struct brw_shader_params
    bool needs_register_pressure;
    void *log_data;
    bool debug_enabled;
+
+   debug_archiver *archiver;
 };
 
 struct brw_shader
@@ -223,6 +225,8 @@ public:
 
    struct brw_shader_stats shader_stats;
 
+   debug_archiver *archiver;
+
    void debug_optimizer(const nir_shader *nir,
                         const char *pass_name,
                         int iteration, int pass_num) const;

src/intel/compiler/meson.build

@@ -159,7 +159,7 @@ libintel_compiler_brw = static_library(
   c_args : [no_override_init_args],
   cpp_args : ['-Werror=vla'],
   gnu_symbol_visibility : 'hidden',
-  dependencies : [idep_nir_headers, idep_mesautil, idep_intel_dev, idep_vtn],
+  dependencies : [idep_nir_headers, idep_mesautil, idep_intel_dev, idep_vtn, idep_mda],
   build_by_default : false,
 )
 

src/intel/dev/intel_debug.c

@@ -71,6 +71,7 @@ static const struct debug_control_bitset debug_control[] = {
    OPT1("blorp",             DEBUG_BLORP),
    OPT1("nodualobj",         DEBUG_NO_DUAL_OBJECT_GS),
    OPT1("optimizer",         DEBUG_OPTIMIZER),
+   OPT1("mda",               DEBUG_MDA),
    OPT1("ann",               DEBUG_ANNOTATION),
    OPT1("no8",               DEBUG_NO8),
    OPT1("no-oaconfig",       DEBUG_NO_OACONFIG),

src/intel/dev/intel_debug.h

@@ -57,6 +57,7 @@ enum intel_debug_flag {
    DEBUG_BLORP,
    DEBUG_NO_DUAL_OBJECT_GS,
    DEBUG_OPTIMIZER,
+   DEBUG_MDA,
    DEBUG_ANNOTATION,
    DEBUG_NO_OACONFIG,
    DEBUG_SPILL_FS,