glsl: add initial implementation of shader cache

[mesa.git] / src / compiler / glsl / shader_cache.cpp

/*
 * Copyright © 2014 Intel Corporation
 *
 * Permission is hereby granted, free of charge, to any person obtaining a
 * copy of this software and associated documentation files (the "Software"),
 * to deal in the Software without restriction, including without limitation
 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
 * and/or sell copies of the Software, and to permit persons to whom the
 * Software is furnished to do so, subject to the following conditions:
 *
 * The above copyright notice and this permission notice (including the next
 * paragraph) shall be included in all copies or substantial portions of the
 * Software.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
 * DEALINGS IN THE SOFTWARE.
 */

/**
 * \file shader_cache.cpp
 *
 * GLSL shader cache implementation
 *
 * This uses disk_cache.c to write out a serialization of various
 * state that's required in order to successfully load and use a
 * binary written out by a drivers backend, this state is referred to as
 * "metadata" throughout the implementation.
 *
 * The hash key for glsl metadata is a hash of the hashes of each GLSL
 * source string as well as some API settings that change the final program
 * such as SSO, attribute bindings, frag data bindings, etc.
 *
 * In order to avoid caching any actual IR we use the put_key/get_key support
 * in the disk_cache to put the SHA-1 hash for each successfully compiled
 * shader into the cache, and optimisticly return early from glCompileShader
 * (if the identical shader had been successfully compiled in the past),
 * in the hope that the final linked shader will be found in the cache.
 * If anything goes wrong (shader variant not found, backend cache item is
 * corrupt, etc) we will use a fallback path to compile and link the IR.
 */

#include "blob.h"
#include "compiler/shader_info.h"
#include "glsl_symbol_table.h"
#include "glsl_parser_extras.h"
#include "ir.h"
#include "ir_optimization.h"
#include "ir_rvalue_visitor.h"
#include "ir_uniform.h"
#include "linker.h"
#include "link_varyings.h"
#include "main/core.h"
#include "nir.h"
#include "program.h"
#include "util/disk_cache.h"
#include "util/mesa-sha1.h"
#include "util/string_to_uint_map.h"

extern "C" {
#include "main/enums.h"
#include "main/shaderobj.h"
#include "program/program.h"
}

static void
compile_shaders(struct gl_context *ctx, struct gl_shader_program *prog) {
   for (unsigned i = 0; i < prog->NumShaders; i++) {
      _mesa_glsl_compile_shader(ctx, prog->Shaders[i], false, false, true);
   }
}

static void
encode_type_to_blob(struct blob *blob, const glsl_type *type)
{
   uint32_t encoding;

   switch (type->base_type) {
   case GLSL_TYPE_UINT:
   case GLSL_TYPE_INT:
   case GLSL_TYPE_FLOAT:
   case GLSL_TYPE_BOOL:
   case GLSL_TYPE_DOUBLE:
   case GLSL_TYPE_UINT64:
   case GLSL_TYPE_INT64:
      encoding = (type->base_type << 24) |
         (type->vector_elements << 4) |
         (type->matrix_columns);
      break;
   case GLSL_TYPE_SAMPLER:
      encoding = (type->base_type) << 24 |
         (type->sampler_dimensionality << 4) |
         (type->sampler_shadow << 3) |
         (type->sampler_array << 2) |
         (type->sampled_type);
      break;
   case GLSL_TYPE_SUBROUTINE:
      encoding = type->base_type << 24;
      blob_write_uint32(blob, encoding);
      blob_write_string(blob, type->name);
      return;
   case GLSL_TYPE_IMAGE:
      encoding = (type->base_type) << 24 |
         (type->sampler_dimensionality << 3) |
         (type->sampler_array << 2) |
         (type->sampled_type);
      break;
   case GLSL_TYPE_ATOMIC_UINT:
      encoding = (type->base_type << 24);
      break;
   case GLSL_TYPE_ARRAY:
      blob_write_uint32(blob, (type->base_type) << 24);
      blob_write_uint32(blob, type->length);
      encode_type_to_blob(blob, type->fields.array);
      return;
   case GLSL_TYPE_STRUCT:
   case GLSL_TYPE_INTERFACE:
      blob_write_uint32(blob, (type->base_type) << 24);
      blob_write_string(blob, type->name);
      blob_write_uint32(blob, type->length);
      blob_write_bytes(blob, type->fields.structure,
                       sizeof(glsl_struct_field) * type->length);
      for (unsigned i = 0; i < type->length; i++) {
         encode_type_to_blob(blob, type->fields.structure[i].type);
         blob_write_string(blob, type->fields.structure[i].name);
      }

      if (type->base_type == GLSL_TYPE_INTERFACE) {
         blob_write_uint32(blob, type->interface_packing);
         blob_write_uint32(blob, type->interface_row_major);
      }
      return;
   case GLSL_TYPE_VOID:
   case GLSL_TYPE_ERROR:
   default:
      assert(!"Cannot encode type!");
      encoding = 0;
      break;
   }

   blob_write_uint32(blob, encoding);
}

static const glsl_type *
decode_type_from_blob(struct blob_reader *blob)
{
   uint32_t u = blob_read_uint32(blob);
   glsl_base_type base_type = (glsl_base_type) (u >> 24);

   switch (base_type) {
   case GLSL_TYPE_UINT:
   case GLSL_TYPE_INT:
   case GLSL_TYPE_FLOAT:
   case GLSL_TYPE_BOOL:
   case GLSL_TYPE_DOUBLE:
   case GLSL_TYPE_UINT64:
   case GLSL_TYPE_INT64:
      return glsl_type::get_instance(base_type, (u >> 4) & 0x0f, u & 0x0f);
   case GLSL_TYPE_SAMPLER:
      return glsl_type::get_sampler_instance((enum glsl_sampler_dim) ((u >> 4) & 0x07),
                                             (u >> 3) & 0x01,
                                             (u >> 2) & 0x01,
                                             (glsl_base_type) ((u >> 0) & 0x03));
   case GLSL_TYPE_SUBROUTINE:
      return glsl_type::get_subroutine_instance(blob_read_string(blob));
   case GLSL_TYPE_IMAGE:
      return glsl_type::get_image_instance((enum glsl_sampler_dim) ((u >> 3) & 0x07),
                                             (u >> 2) & 0x01,
                                             (glsl_base_type) ((u >> 0) & 0x03));
   case GLSL_TYPE_ATOMIC_UINT:
      return glsl_type::atomic_uint_type;
   case GLSL_TYPE_ARRAY: {
      unsigned length = blob_read_uint32(blob);
      return glsl_type::get_array_instance(decode_type_from_blob(blob),
                                           length);
   }
   case GLSL_TYPE_STRUCT:
   case GLSL_TYPE_INTERFACE: {
      char *name = blob_read_string(blob);
      unsigned num_fields = blob_read_uint32(blob);
      glsl_struct_field *fields = (glsl_struct_field *)
         blob_read_bytes(blob, sizeof(glsl_struct_field) * num_fields);
      for (unsigned i = 0; i < num_fields; i++) {
         fields[i].type = decode_type_from_blob(blob);
         fields[i].name = blob_read_string(blob);
      }

      if (base_type == GLSL_TYPE_INTERFACE) {
         enum glsl_interface_packing packing =
            (glsl_interface_packing) blob_read_uint32(blob);
         bool row_major = blob_read_uint32(blob);
         return glsl_type::get_interface_instance(fields, num_fields,
                                                  packing, row_major, name);
      } else {
         return glsl_type::get_record_instance(fields, num_fields, name);
      }
   }
   case GLSL_TYPE_VOID:
   case GLSL_TYPE_ERROR:
   default:
      assert(!"Cannot decode type!");
      return NULL;
   }
}

static void
write_uniforms(struct blob *metadata, struct gl_shader_program *prog)
{
   blob_write_uint32(metadata, prog->SamplersValidated);
   blob_write_uint32(metadata, prog->data->NumUniformStorage);
   blob_write_uint32(metadata, prog->data->NumUniformDataSlots);

   for (unsigned i = 0; i < prog->data->NumUniformStorage; i++) {
      encode_type_to_blob(metadata, prog->data->UniformStorage[i].type);
      blob_write_uint32(metadata, prog->data->UniformStorage[i].array_elements);
      blob_write_string(metadata, prog->data->UniformStorage[i].name);
      blob_write_uint32(metadata, prog->data->UniformStorage[i].storage -
                                  prog->data->UniformDataSlots);
      blob_write_uint32(metadata, prog->data->UniformStorage[i].remap_location);
      blob_write_uint32(metadata, prog->data->UniformStorage[i].block_index);
      blob_write_uint32(metadata, prog->data->UniformStorage[i].atomic_buffer_index);
      blob_write_uint32(metadata, prog->data->UniformStorage[i].offset);
      blob_write_uint32(metadata, prog->data->UniformStorage[i].array_stride);
      blob_write_uint32(metadata, prog->data->UniformStorage[i].matrix_stride);
      blob_write_uint32(metadata, prog->data->UniformStorage[i].row_major);
      blob_write_uint32(metadata,
                        prog->data->UniformStorage[i].num_compatible_subroutines);
      blob_write_uint32(metadata,
                        prog->data->UniformStorage[i].top_level_array_size);
      blob_write_uint32(metadata,
                        prog->data->UniformStorage[i].top_level_array_stride);
   }
}

static void
read_uniforms(struct blob_reader *metadata, struct gl_shader_program *prog)
{
   struct gl_uniform_storage *uniforms;
   union gl_constant_value *data;

   prog->SamplersValidated = blob_read_uint32(metadata);
   prog->data->NumUniformStorage = blob_read_uint32(metadata);
   prog->data->NumUniformDataSlots = blob_read_uint32(metadata);

   uniforms = rzalloc_array(prog, struct gl_uniform_storage,
                            prog->data->NumUniformStorage);
   prog->data->UniformStorage = uniforms;

   data = rzalloc_array(uniforms, union gl_constant_value,
                        prog->data->NumUniformDataSlots);
   prog->data->UniformDataSlots = data;

   prog->UniformHash = new string_to_uint_map;

   for (unsigned i = 0; i < prog->data->NumUniformStorage; i++) {
      uniforms[i].type = decode_type_from_blob(metadata);
      uniforms[i].array_elements = blob_read_uint32(metadata);
      uniforms[i].name = ralloc_strdup(prog, blob_read_string (metadata));
      uniforms[i].storage = data + blob_read_uint32(metadata);
      uniforms[i].remap_location = blob_read_uint32(metadata);
      uniforms[i].block_index = blob_read_uint32(metadata);
      uniforms[i].atomic_buffer_index = blob_read_uint32(metadata);
      uniforms[i].offset = blob_read_uint32(metadata);
      uniforms[i].array_stride = blob_read_uint32(metadata);
      uniforms[i].matrix_stride = blob_read_uint32(metadata);
      uniforms[i].row_major = blob_read_uint32(metadata);
      uniforms[i].num_compatible_subroutines = blob_read_uint32(metadata);
      uniforms[i].top_level_array_size = blob_read_uint32(metadata);
      uniforms[i].top_level_array_stride = blob_read_uint32(metadata);
      prog->UniformHash->put(i, uniforms[i].name);
   }
}


static void
write_uniform_remap_table(struct blob *metadata,
                          struct gl_shader_program *prog)
{
   blob_write_uint32(metadata, prog->NumUniformRemapTable);

   for (unsigned i = 0; i < prog->NumUniformRemapTable; i++) {
      blob_write_uint32(metadata, prog->UniformRemapTable[i] -
                           prog->data->UniformStorage);
   }
}

static void
read_uniform_remap_table(struct blob_reader *metadata,
                         struct gl_shader_program *prog)
{
   prog->NumUniformRemapTable = blob_read_uint32(metadata);

   prog->UniformRemapTable =rzalloc_array(prog, struct gl_uniform_storage *,
                                          prog->NumUniformRemapTable);

   for (unsigned i = 0; i < prog->NumUniformRemapTable; i++) {
      prog->UniformRemapTable[i] =
         prog->data->UniformStorage + blob_read_uint32(metadata);
   }
}

static void
write_shader_parameters(struct blob *metadata,
                        struct gl_program_parameter_list *params)
{
   blob_write_uint32(metadata, params->NumParameters);
   uint32_t i = 0;

   while (i < params->NumParameters) {
      struct gl_program_parameter *param = &params->Parameters[i];

      blob_write_uint32(metadata, param->Type);
      blob_write_string(metadata, param->Name);
      blob_write_uint32(metadata, param->Size);
      blob_write_uint32(metadata, param->DataType);
      blob_write_bytes(metadata, param->StateIndexes,
                       sizeof(param->StateIndexes));

      i += (param->Size + 3) / 4;
   }

   blob_write_bytes(metadata, params->ParameterValues,
                    sizeof(gl_constant_value) * 4 * params->NumParameters);

   blob_write_uint32(metadata, params->StateFlags);
}

static void
read_shader_parameters(struct blob_reader *metadata,
                       struct gl_program_parameter_list *params)
{
   gl_state_index state_indexes[STATE_LENGTH];
   uint32_t i = 0;
   uint32_t num_parameters = blob_read_uint32(metadata);

   while (i < num_parameters) {
      gl_register_file type = (gl_register_file) blob_read_uint32(metadata);
      const char *name = blob_read_string(metadata);
      unsigned size = blob_read_uint32(metadata);
      unsigned data_type = blob_read_uint32(metadata);
      blob_copy_bytes(metadata, (uint8_t *) state_indexes,
                      sizeof(state_indexes));

      _mesa_add_parameter(params, type, name, size, data_type,
                          NULL, state_indexes);

      i += (size + 3) / 4;
   }

   blob_copy_bytes(metadata, (uint8_t *) params->ParameterValues,
                    sizeof(gl_constant_value) * 4 * params->NumParameters);

   params->StateFlags = blob_read_uint32(metadata);
}

static void
write_shader_metadata(struct blob *metadata, gl_linked_shader *shader)
{
   assert(shader->Program);
   struct gl_program *glprog = shader->Program;

   blob_write_bytes(metadata, glprog->TexturesUsed,
                    sizeof(glprog->TexturesUsed));
   blob_write_uint64(metadata, glprog->SamplersUsed);

   write_shader_parameters(metadata, glprog->Parameters);
}

static void
read_shader_metadata(struct blob_reader *metadata,
                     struct gl_program *glprog,
                     gl_linked_shader *linked)
{
   blob_copy_bytes(metadata, (uint8_t *) glprog->TexturesUsed,
                   sizeof(glprog->TexturesUsed));
   glprog->SamplersUsed = blob_read_uint64(metadata);

   glprog->Parameters = _mesa_new_parameter_list();
   read_shader_parameters(metadata, glprog->Parameters);
}

static void
create_binding_str(const char *key, unsigned value, void *closure)
{
   char **bindings_str = (char **) closure;
   ralloc_asprintf_append(bindings_str, "%s:%u,", key, value);
}

static void
create_linked_shader_and_program(struct gl_context *ctx,
                                 gl_shader_stage stage,
                                 struct gl_shader_program *prog,
                                 struct blob_reader *metadata)
{
   struct gl_program *glprog;

   struct gl_linked_shader *linked = rzalloc(NULL, struct gl_linked_shader);
   linked->Stage = stage;

   glprog = ctx->Driver.NewProgram(ctx, _mesa_shader_stage_to_program(stage),
                                   prog->Name, false);
   glprog->info.stage = stage;
   linked->Program = glprog;

   read_shader_metadata(metadata, glprog, linked);

   /* Restore shader info */
   blob_copy_bytes(metadata, (uint8_t *) &glprog->info, sizeof(shader_info));
   if (glprog->info.name)
      glprog->info.name = ralloc_strdup(glprog, blob_read_string(metadata));
   if (glprog->info.label)
      glprog->info.label = ralloc_strdup(glprog, blob_read_string(metadata));

   _mesa_reference_shader_program_data(ctx, &glprog->sh.data, prog->data);
   _mesa_reference_program(ctx, &linked->Program, glprog);
   prog->_LinkedShaders[stage] = linked;
}

void
shader_cache_write_program_metadata(struct gl_context *ctx,
                                    struct gl_shader_program *prog)
{
   struct disk_cache *cache = ctx->Cache;
   if (!cache)
      return;

   /* Exit early when we are dealing with a ff shader with no source file to
    * generate a source from.
    *
    * TODO: In future we should use another method to generate a key for ff
    * programs.
    */
   if (*prog->data->sha1 == 0)
      return;

   struct blob *metadata = blob_create(NULL);

   write_uniforms(metadata, prog);

   blob_write_uint32(metadata, prog->data->Version);
   blob_write_uint32(metadata, prog->data->linked_stages);

   for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
      struct gl_linked_shader *sh = prog->_LinkedShaders[i];
      if (sh) {
         write_shader_metadata(metadata, sh);

         /* Store nir shader info */
         blob_write_bytes(metadata, &sh->Program->info, sizeof(shader_info));

         if (sh->Program->info.name)
            blob_write_string(metadata, sh->Program->info.name);

         if (sh->Program->info.label)
            blob_write_string(metadata, sh->Program->info.label);
      }
   }

   write_uniform_remap_table(metadata, prog);

   char sha1_buf[41];
   for (unsigned i = 0; i < prog->NumShaders; i++) {
      disk_cache_put_key(cache, prog->Shaders[i]->sha1);
      if (ctx->_Shader->Flags & GLSL_CACHE_INFO) {
         fprintf(stderr, "marking shader: %s\n",
                 _mesa_sha1_format(sha1_buf, prog->Shaders[i]->sha1));
      }
   }

   disk_cache_put(cache, prog->data->sha1, metadata->data, metadata->size);

   ralloc_free(metadata);

   if (ctx->_Shader->Flags & GLSL_CACHE_INFO) {
      fprintf(stderr, "putting program metadata in cache: %s\n",
              _mesa_sha1_format(sha1_buf, prog->data->sha1));
   }
}

bool
shader_cache_read_program_metadata(struct gl_context *ctx,
                                   struct gl_shader_program *prog)
{
   /* Fixed function programs generated by Mesa are not cached. So don't
    * try to read metadata for them from the cache.
    */
   if (prog->Name == 0)
      return false;

   struct disk_cache *cache = ctx->Cache;
   if (!cache)
      return false;

   /* Include bindings when creating sha1. These bindings change the resulting
    * binary so they are just as important as the shader source.
    */
   char *buf = ralloc_strdup(NULL, "vb: ");
   prog->AttributeBindings->iterate(create_binding_str, &buf);
   ralloc_strcat(&buf, "fb: ");
   prog->FragDataBindings->iterate(create_binding_str, &buf);
   ralloc_strcat(&buf, "fbi: ");
   prog->FragDataIndexBindings->iterate(create_binding_str, &buf);

   /* SSO has an effect on the linked program so include this when generating
    * the sha also.
    */
   ralloc_asprintf_append(&buf, "sso: %s\n",
                          prog->SeparateShader ? "T" : "F");

   char sha1buf[41];
   for (unsigned i = 0; i < prog->NumShaders; i++) {
      struct gl_shader *sh = prog->Shaders[i];
      ralloc_asprintf_append(&buf, "%s: %s\n",
                             _mesa_shader_stage_to_abbrev(sh->Stage),
                             _mesa_sha1_format(sha1buf, sh->sha1));
   }
   _mesa_sha1_compute(buf, strlen(buf), prog->data->sha1);
   ralloc_free(buf);

   size_t size;
   uint8_t *buffer = (uint8_t *) disk_cache_get(cache, prog->data->sha1,
                                                &size);
   if (buffer == NULL) {
      /* Cached program not found. We may have seen the individual shaders
       * before and skipped compiling but they may not have been used together
       * in this combination before. Fall back to linking shaders but first
       * re-compile the shaders.
       *
       * We could probably only compile the shaders which were skipped here
       * but we need to be careful because the source may also have been
       * changed since the last compile so for now we just recompile
       * everything.
       */
      compile_shaders(ctx, prog);
      return false;
   }

   if (ctx->_Shader->Flags & GLSL_CACHE_INFO) {
      fprintf(stderr, "loading shader program meta data from cache: %s\n",
              _mesa_sha1_format(sha1buf, prog->data->sha1));
   }

   struct blob_reader metadata;
   blob_reader_init(&metadata, buffer, size);

   assert(prog->data->UniformStorage == NULL);

   read_uniforms(&metadata, prog);

   prog->data->Version = blob_read_uint32(&metadata);
   prog->data->linked_stages = blob_read_uint32(&metadata);

   unsigned mask = prog->data->linked_stages;
   while (mask) {
      const int j = u_bit_scan(&mask);
      create_linked_shader_and_program(ctx, (gl_shader_stage) j, prog,
                                       &metadata);
   }

   read_uniform_remap_table(&metadata, prog);

   if (metadata.current != metadata.end || metadata.overrun) {
      /* Something has gone wrong discard the item from the cache and rebuild
       * from source.
       */
      assert(!"Invalid GLSL shader disk cache item!");

      if (ctx->_Shader->Flags & GLSL_CACHE_INFO) {
         fprintf(stderr, "Error reading program from cache (invalid GLSL "
                 "cache item)\n");
      }

      disk_cache_remove(cache, prog->data->sha1);
      compile_shaders(ctx, prog);
      free(buffer);
      return false;
   }

   /* This is used to flag a shader retrieved from cache */
   prog->data->LinkStatus = linking_skipped;

   free (buffer);

   return true;
}