X-Git-Url: https://git.libre-soc.org/?a=blobdiff_plain;f=src%2Fmesa%2Fstate_tracker%2Fst_shader_cache.c;h=344bd93d6671c5357a0f60b16e27c13861bcf926;hb=33de483d55757024baafca9a69c2159a0a61481f;hp=62d62f761177294d5d661ec052f80c06d44915e0;hpb=9eebc55cc206d7400cc33e5faa52b51a1b4b5b8b;p=mesa.git

diff --git a/src/mesa/state_tracker/st_shader_cache.c b/src/mesa/state_tracker/st_shader_cache.c
index 62d62f76117..344bd93d667 100644
--- a/src/mesa/state_tracker/st_shader_cache.c
+++ b/src/mesa/state_tracker/st_shader_cache.c
@@ -25,11 +25,20 @@
 #include "st_debug.h"
 #include "st_program.h"
 #include "st_shader_cache.h"
+#include "st_util.h"
 #include "compiler/glsl/program.h"
+#include "compiler/nir/nir.h"
+#include "compiler/nir/nir_serialize.h"
 #include "pipe/p_shader_tokens.h"
 #include "program/ir_to_mesa.h"
 #include "util/u_memory.h"
 
+void
+st_get_program_binary_driver_sha1(struct gl_context *ctx, uint8_t *sha1)
+{
+   disk_cache_compute_key(ctx->Cache, NULL, 0, sha1);
+}
+
 static void
 write_stream_out_to_cache(struct blob *blob,
                           struct pipe_shader_state *tgsi)
@@ -38,21 +47,37 @@ write_stream_out_to_cache(struct blob *blob,
                     sizeof(tgsi->stream_output));
 }
 
+static void
+copy_blob_to_driver_cache_blob(struct blob *blob, struct gl_program *prog)
+{
+   prog->driver_cache_blob = ralloc_size(NULL, blob->size);
+   memcpy(prog->driver_cache_blob, blob->data, blob->size);
+   prog->driver_cache_blob_size = blob->size;
+}
+
 static void
 write_tgsi_to_cache(struct blob *blob, const struct tgsi_token *tokens,
                     struct gl_program *prog, unsigned num_tokens)
 {
    blob_write_uint32(blob, num_tokens);
    blob_write_bytes(blob, tokens, num_tokens * sizeof(struct tgsi_token));
+   copy_blob_to_driver_cache_blob(blob, prog);
+}
 
-   prog->driver_cache_blob = ralloc_size(NULL, blob->size);
-   memcpy(prog->driver_cache_blob, blob->data, blob->size);
-   prog->driver_cache_blob_size = blob->size;
+static void
+write_nir_to_cache(struct blob *blob, struct gl_program *prog)
+{
+   nir_serialize(blob, prog->nir, false);
+   copy_blob_to_driver_cache_blob(blob, prog);
 }
 
-void
-st_serialise_tgsi_program(struct gl_context *ctx, struct gl_program *prog)
+static void
+st_serialise_ir_program(struct gl_context *ctx, struct gl_program *prog,
+                        bool nir)
 {
+   if (prog->driver_cache_blob)
+      return;
+
    struct blob blob;
    blob_init(&blob);
 
@@ -63,12 +88,18 @@ st_serialise_tgsi_program(struct gl_context *ctx, struct gl_program *prog)
       blob_write_uint32(&blob, stvp->num_inputs);
       blob_write_bytes(&blob, stvp->index_to_input,
                        sizeof(stvp->index_to_input));
+      blob_write_bytes(&blob, stvp->input_to_index,
+                       sizeof(stvp->input_to_index));
       blob_write_bytes(&blob, stvp->result_to_output,
                        sizeof(stvp->result_to_output));
 
       write_stream_out_to_cache(&blob, &stvp->tgsi);
-      write_tgsi_to_cache(&blob, stvp->tgsi.tokens, prog,
-                          stvp->num_tgsi_tokens);
+
+      if (nir)
+         write_nir_to_cache(&blob, prog);
+      else
+         write_tgsi_to_cache(&blob, stvp->tgsi.tokens, prog,
+                             stvp->num_tgsi_tokens);
       break;
    }
    case MESA_SHADER_TESS_CTRL:
@@ -77,22 +108,32 @@ st_serialise_tgsi_program(struct gl_context *ctx, struct gl_program *prog)
       struct st_common_program *stcp = (struct st_common_program *) prog;
 
       write_stream_out_to_cache(&blob, &stcp->tgsi);
-      write_tgsi_to_cache(&blob, stcp->tgsi.tokens, prog,
-                          stcp->num_tgsi_tokens);
+
+      if (nir)
+         write_nir_to_cache(&blob, prog);
+      else
+         write_tgsi_to_cache(&blob, stcp->tgsi.tokens, prog,
+                             stcp->num_tgsi_tokens);
       break;
    }
    case MESA_SHADER_FRAGMENT: {
       struct st_fragment_program *stfp = (struct st_fragment_program *) prog;
 
-      write_tgsi_to_cache(&blob, stfp->tgsi.tokens, prog,
-                          stfp->num_tgsi_tokens);
+      if (nir)
+         write_nir_to_cache(&blob, prog);
+      else
+         write_tgsi_to_cache(&blob, stfp->tgsi.tokens, prog,
+                             stfp->num_tgsi_tokens);
       break;
    }
    case MESA_SHADER_COMPUTE: {
       struct st_compute_program *stcp = (struct st_compute_program *) prog;
 
-      write_tgsi_to_cache(&blob, stcp->tgsi.prog, prog,
-                          stcp->num_tgsi_tokens);
+      if (nir)
+         write_nir_to_cache(&blob, prog);
+      else
+         write_tgsi_to_cache(&blob, stcp->tgsi.prog, prog,
+                             stcp->num_tgsi_tokens);
       break;
    }
    default:
@@ -106,7 +147,8 @@ st_serialise_tgsi_program(struct gl_context *ctx, struct gl_program *prog)
  * Store tgsi and any other required state in on-disk shader cache.
  */
 void
-st_store_tgsi_in_disk_cache(struct st_context *st, struct gl_program *prog)
+st_store_ir_in_disk_cache(struct st_context *st, struct gl_program *prog,
+                          bool nir)
 {
    if (!st->ctx->Cache)
       return;
@@ -118,10 +160,10 @@ st_store_tgsi_in_disk_cache(struct st_context *st, struct gl_program *prog)
    if (memcmp(prog->sh.data->sha1, zero, sizeof(prog->sh.data->sha1)) == 0)
       return;
 
-   st_serialise_tgsi_program(st->ctx, prog);
+   st_serialise_ir_program(st->ctx, prog, nir);
 
    if (st->ctx->_Shader->Flags & GLSL_CACHE_INFO) {
-      fprintf(stderr, "putting %s tgsi_tokens in cache\n",
+      fprintf(stderr, "putting %s state tracker IR in cache\n",
               _mesa_shader_stage_to_string(prog->info.stage));
    }
 }
@@ -136,22 +178,27 @@ read_stream_out_from_cache(struct blob_reader *blob_reader,
 
 static void
 read_tgsi_from_cache(struct blob_reader *blob_reader,
-                     const struct tgsi_token **tokens)
+                     const struct tgsi_token **tokens,
+                     unsigned *num_tokens)
 {
-   uint32_t num_tokens  = blob_read_uint32(blob_reader);
-   unsigned tokens_size = num_tokens * sizeof(struct tgsi_token);
+   *num_tokens  = blob_read_uint32(blob_reader);
+   unsigned tokens_size = *num_tokens * sizeof(struct tgsi_token);
    *tokens = (const struct tgsi_token*) MALLOC(tokens_size);
    blob_copy_bytes(blob_reader, (uint8_t *) *tokens, tokens_size);
 }
 
-void
-st_deserialise_tgsi_program(struct gl_context *ctx,
-                            struct gl_shader_program *shProg,
-                            struct gl_program *prog)
+static void
+st_deserialise_ir_program(struct gl_context *ctx,
+                          struct gl_shader_program *shProg,
+                          struct gl_program *prog, bool nir)
 {
    struct st_context *st = st_context(ctx);
    size_t size = prog->driver_cache_blob_size;
    uint8_t *buffer = (uint8_t *) prog->driver_cache_blob;
+   const struct nir_shader_compiler_options *options =
+      ctx->Const.ShaderCompilerOptions[prog->info.stage].NirOptions;
+
+   assert(prog->driver_cache_blob && prog->driver_cache_blob_size > 0);
 
    struct blob_reader blob_reader;
    blob_reader_init(&blob_reader, buffer, size);
@@ -165,11 +212,22 @@ st_deserialise_tgsi_program(struct gl_context *ctx,
       stvp->num_inputs = blob_read_uint32(&blob_reader);
       blob_copy_bytes(&blob_reader, (uint8_t *) stvp->index_to_input,
                       sizeof(stvp->index_to_input));
+      blob_copy_bytes(&blob_reader, (uint8_t *) stvp->input_to_index,
+                      sizeof(stvp->input_to_index));
       blob_copy_bytes(&blob_reader, (uint8_t *) stvp->result_to_output,
                       sizeof(stvp->result_to_output));
 
       read_stream_out_from_cache(&blob_reader, &stvp->tgsi);
-      read_tgsi_from_cache(&blob_reader, &stvp->tgsi.tokens);
+
+      if (nir) {
+         stvp->tgsi.type = PIPE_SHADER_IR_NIR;
+         stvp->shader_program = shProg;
+         stvp->tgsi.ir.nir = nir_deserialize(NULL, options, &blob_reader);
+         prog->nir = stvp->tgsi.ir.nir;
+      } else {
+         read_tgsi_from_cache(&blob_reader, &stvp->tgsi.tokens,
+                              &stvp->num_tgsi_tokens);
+      }
 
       if (st->vp == stvp)
          st->dirty |= ST_NEW_VERTEX_PROGRAM(st, stvp);
@@ -179,11 +237,18 @@ st_deserialise_tgsi_program(struct gl_context *ctx,
    case MESA_SHADER_TESS_CTRL: {
       struct st_common_program *sttcp = st_common_program(prog);
 
-      st_release_basic_variants(st, sttcp->Base.Target,
-                                &sttcp->variants, &sttcp->tgsi);
-
+      st_release_basic_variants(st, sttcp);
       read_stream_out_from_cache(&blob_reader, &sttcp->tgsi);
-      read_tgsi_from_cache(&blob_reader, &sttcp->tgsi.tokens);
+
+      if (nir) {
+         sttcp->tgsi.type = PIPE_SHADER_IR_NIR;
+         sttcp->shader_program = shProg;
+         sttcp->tgsi.ir.nir = nir_deserialize(NULL, options, &blob_reader);
+         prog->nir = sttcp->tgsi.ir.nir;
+      } else {
+         read_tgsi_from_cache(&blob_reader, &sttcp->tgsi.tokens,
+                              &sttcp->num_tgsi_tokens);
+      }
 
       if (st->tcp == sttcp)
          st->dirty |= sttcp->affected_states;
@@ -193,11 +258,18 @@ st_deserialise_tgsi_program(struct gl_context *ctx,
    case MESA_SHADER_TESS_EVAL: {
       struct st_common_program *sttep = st_common_program(prog);
 
-      st_release_basic_variants(st, sttep->Base.Target,
-                                &sttep->variants, &sttep->tgsi);
-
+      st_release_basic_variants(st, sttep);
       read_stream_out_from_cache(&blob_reader, &sttep->tgsi);
-      read_tgsi_from_cache(&blob_reader, &sttep->tgsi.tokens);
+
+      if (nir) {
+         sttep->tgsi.type = PIPE_SHADER_IR_NIR;
+         sttep->shader_program = shProg;
+         sttep->tgsi.ir.nir = nir_deserialize(NULL, options, &blob_reader);
+         prog->nir = sttep->tgsi.ir.nir;
+      } else {
+         read_tgsi_from_cache(&blob_reader, &sttep->tgsi.tokens,
+                              &sttep->num_tgsi_tokens);
+      }
 
       if (st->tep == sttep)
          st->dirty |= sttep->affected_states;
@@ -207,11 +279,18 @@ st_deserialise_tgsi_program(struct gl_context *ctx,
    case MESA_SHADER_GEOMETRY: {
       struct st_common_program *stgp = st_common_program(prog);
 
-      st_release_basic_variants(st, stgp->Base.Target, &stgp->variants,
-                                &stgp->tgsi);
-
+      st_release_basic_variants(st, stgp);
       read_stream_out_from_cache(&blob_reader, &stgp->tgsi);
-      read_tgsi_from_cache(&blob_reader, &stgp->tgsi.tokens);
+
+      if (nir) {
+         stgp->tgsi.type = PIPE_SHADER_IR_NIR;
+         stgp->shader_program = shProg;
+         stgp->tgsi.ir.nir = nir_deserialize(NULL, options, &blob_reader);
+         prog->nir = stgp->tgsi.ir.nir;
+      } else {
+         read_tgsi_from_cache(&blob_reader, &stgp->tgsi.tokens,
+                              &stgp->num_tgsi_tokens);
+      }
 
       if (st->gp == stgp)
          st->dirty |= stgp->affected_states;
@@ -223,7 +302,15 @@ st_deserialise_tgsi_program(struct gl_context *ctx,
 
       st_release_fp_variants(st, stfp);
 
-      read_tgsi_from_cache(&blob_reader, &stfp->tgsi.tokens);
+      if (nir) {
+         stfp->tgsi.type = PIPE_SHADER_IR_NIR;
+         stfp->shader_program = shProg;
+         stfp->tgsi.ir.nir = nir_deserialize(NULL, options, &blob_reader);
+         prog->nir = stfp->tgsi.ir.nir;
+      } else {
+         read_tgsi_from_cache(&blob_reader, &stfp->tgsi.tokens,
+                              &stfp->num_tgsi_tokens);
+      }
 
       if (st->fp == stfp)
          st->dirty |= stfp->affected_states;
@@ -235,8 +322,16 @@ st_deserialise_tgsi_program(struct gl_context *ctx,
 
       st_release_cp_variants(st, stcp);
 
-      read_tgsi_from_cache(&blob_reader,
-                           (const struct tgsi_token**) &stcp->tgsi.prog);
+      if (nir) {
+         stcp->tgsi.ir_type = PIPE_SHADER_IR_NIR;
+         stcp->shader_program = shProg;
+         stcp->tgsi.prog = nir_deserialize(NULL, options, &blob_reader);
+         prog->nir = (nir_shader *) stcp->tgsi.prog;
+      } else {
+         read_tgsi_from_cache(&blob_reader,
+                              (const struct tgsi_token**) &stcp->tgsi.prog,
+                              &stcp->num_tgsi_tokens);
+      }
 
       stcp->tgsi.req_local_mem = stcp->Base.info.cs.shared_size;
       stcp->tgsi.req_private_mem = 0;
@@ -265,7 +360,7 @@ st_deserialise_tgsi_program(struct gl_context *ctx,
    }
 
    st_set_prog_affected_state_flags(prog);
-   _mesa_associate_uniform_storage(ctx, shProg, prog, false);
+   _mesa_associate_uniform_storage(ctx, shProg, prog);
 
    /* Create Gallium shaders now instead of on demand. */
    if (ST_DEBUG & DEBUG_PRECOMPILE ||
@@ -274,8 +369,9 @@ st_deserialise_tgsi_program(struct gl_context *ctx,
 }
 
 bool
-st_load_tgsi_from_disk_cache(struct gl_context *ctx,
-                             struct gl_shader_program *prog)
+st_load_ir_from_disk_cache(struct gl_context *ctx,
+                           struct gl_shader_program *prog,
+                           bool nir)
 {
    if (!ctx->Cache)
       return false;
@@ -283,7 +379,7 @@ st_load_tgsi_from_disk_cache(struct gl_context *ctx,
    /* If we didn't load the GLSL metadata from cache then we could not have
     * loaded the tgsi either.
     */
-   if (prog->data->LinkStatus != linking_skipped)
+   if (prog->data->LinkStatus != LINKING_SKIPPED)
       return false;
 
    for (unsigned i = 0; i < MESA_SHADER_STAGES; i++) {
@@ -291,7 +387,7 @@ st_load_tgsi_from_disk_cache(struct gl_context *ctx,
          continue;
 
       struct gl_program *glprog = prog->_LinkedShaders[i]->Program;
-      st_deserialise_tgsi_program(ctx, prog, glprog);
+      st_deserialise_ir_program(ctx, prog, glprog, nir);
 
       /* We don't need the cached blob anymore so free it */
       ralloc_free(glprog->driver_cache_blob);
@@ -299,10 +395,54 @@ st_load_tgsi_from_disk_cache(struct gl_context *ctx,
       glprog->driver_cache_blob_size = 0;
 
       if (ctx->_Shader->Flags & GLSL_CACHE_INFO) {
-         fprintf(stderr, "%s tgsi_tokens retrieved from cache\n",
+         fprintf(stderr, "%s state tracker IR retrieved from cache\n",
                  _mesa_shader_stage_to_string(i));
       }
    }
 
    return true;
 }
+
+void
+st_serialise_tgsi_program(struct gl_context *ctx, struct gl_program *prog)
+{
+   st_serialise_ir_program(ctx, prog, false);
+}
+
+void
+st_serialise_tgsi_program_binary(struct gl_context *ctx,
+                                 struct gl_shader_program *shProg,
+                                 struct gl_program *prog)
+{
+   st_serialise_ir_program(ctx, prog, false);
+}
+
+void
+st_deserialise_tgsi_program(struct gl_context *ctx,
+                            struct gl_shader_program *shProg,
+                            struct gl_program *prog)
+{
+   st_deserialise_ir_program(ctx, shProg, prog, false);
+}
+
+void
+st_serialise_nir_program(struct gl_context *ctx, struct gl_program *prog)
+{
+   st_serialise_ir_program(ctx, prog, true);
+}
+
+void
+st_serialise_nir_program_binary(struct gl_context *ctx,
+                                struct gl_shader_program *shProg,
+                                struct gl_program *prog)
+{
+   st_serialise_ir_program(ctx, prog, true);
+}
+
+void
+st_deserialise_nir_program(struct gl_context *ctx,
+                           struct gl_shader_program *shProg,
+                           struct gl_program *prog)
+{
+   st_deserialise_ir_program(ctx, shProg, prog, true);
+}