cso: don't track the number of sampler states bound

This removes 2 loops from hot codepaths and adds 1 loop to a rare codepath
(restore_sampler_states), and makes sanitize_hash() slightly worse.

Sampler states, when bound, are not unbound for draw calls that don't need
them. That's OK, because bound sampler states don't add any overhead.

This results in lower CPU overhead in most cases.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>

diff --git a/src/gallium/auxiliary/cso_cache/cso_context.c b/src/gallium/auxiliary/cso_cache/cso_context.c
index 5558385e4c21f415c6871413dabb82028b4ba039..4947b8e1624531249bd34940a6a0c101eccdb99f 100644 (file)
--- a/src/gallium/auxiliary/cso_cache/cso_context.c
+++ b/src/gallium/auxiliary/cso_cache/cso_context.c
@@ -57,7 +57,6 @@ struct sampler_info
 {
    struct cso_sampler *cso_samplers[PIPE_MAX_SAMPLERS];
    void *samplers[PIPE_MAX_SAMPLERS];
-   unsigned nr_samplers;
 };
 
 
@@ -83,6 +82,11 @@ struct cso_context {
    struct sampler_info fragment_samplers_saved;
    struct sampler_info samplers[PIPE_SHADER_TYPES];
 
+   /* Temporary number until cso_single_sampler_done is called.
+    * It tracks the highest sampler seen in cso_single_sampler.
+    */
+   int max_sampler_seen;
+
    struct pipe_vertex_buffer aux_vertex_buffer_current;
    struct pipe_vertex_buffer aux_vertex_buffer_saved;
    unsigned aux_vertex_buffer_index;
@@ -240,7 +244,7 @@ sanitize_hash(struct cso_hash *hash, enum cso_cache_type type,
        * table, to prevent them from being deleted
        */
       for (i = 0; i < PIPE_SHADER_TYPES; i++) {
-         for (j = 0; j < ctx->samplers[i].nr_samplers; j++) {
+         for (j = 0; j < PIPE_MAX_SAMPLERS; j++) {
             struct cso_sampler *sampler = ctx->samplers[i].cso_samplers[j];
 
             if (sampler && cso_hash_take(hash, sampler->hash_key))
@@ -334,6 +338,7 @@ cso_create_context(struct pipe_context *pipe, unsigned u_vbuf_flags)
       ctx->has_streamout = TRUE;
    }
 
+   ctx->max_sampler_seen = -1;
    return ctx;
 
 out:
@@ -1230,9 +1235,7 @@ cso_single_sampler(struct cso_context *ctx, enum pipe_shader_type shader_stage,
 
       ctx->samplers[shader_stage].cso_samplers[idx] = cso;
       ctx->samplers[shader_stage].samplers[idx] = cso->data;
-   } else {
-      ctx->samplers[shader_stage].cso_samplers[idx] = NULL;
-      ctx->samplers[shader_stage].samplers[idx] = NULL;
+      ctx->max_sampler_seen = MAX2(ctx->max_sampler_seen, (int)idx);
    }
 
    return PIPE_OK;
@@ -1247,19 +1250,14 @@ cso_single_sampler_done(struct cso_context *ctx,
                         enum pipe_shader_type shader_stage)
 {
    struct sampler_info *info = &ctx->samplers[shader_stage];
-   const unsigned old_nr_samplers = info->nr_samplers;
-   unsigned i;
 
-   /* find highest non-null sampler */
-   for (i = PIPE_MAX_SAMPLERS; i > 0; i--) {
-      if (info->samplers[i - 1] != NULL)
-         break;
-   }
+   if (ctx->max_sampler_seen == -1)
+      return;
 
-   info->nr_samplers = i;
    ctx->pipe->bind_sampler_states(ctx->pipe, shader_stage, 0,
-                                  MAX2(old_nr_samplers, info->nr_samplers),
+                                  ctx->max_sampler_seen + 1,
                                   info->samplers);
+   ctx->max_sampler_seen = -1;
 }
 
 
@@ -1274,7 +1272,6 @@ cso_set_samplers(struct cso_context *ctx,
                  unsigned nr,
                  const struct pipe_sampler_state **templates)
 {
-   struct sampler_info *info = &ctx->samplers[shader_stage];
    unsigned i;
    enum pipe_error temp, error = PIPE_OK;
 
@@ -1284,12 +1281,6 @@ cso_set_samplers(struct cso_context *ctx,
          error = temp;
    }
 
-   for ( ; i < info->nr_samplers; i++) {
-      temp = cso_single_sampler(ctx, shader_stage, i, NULL);
-      if (temp != PIPE_OK)
-         error = temp;
-   }
-
    cso_single_sampler_done(ctx, shader_stage);
 
    return error;
@@ -1301,11 +1292,9 @@ cso_save_fragment_samplers(struct cso_context *ctx)
    struct sampler_info *info = &ctx->samplers[PIPE_SHADER_FRAGMENT];
    struct sampler_info *saved = &ctx->fragment_samplers_saved;
 
-   saved->nr_samplers = info->nr_samplers;
-   memcpy(saved->cso_samplers, info->cso_samplers, info->nr_samplers *
-          sizeof(*info->cso_samplers));
-   memcpy(saved->samplers, info->samplers, info->nr_samplers *
-          sizeof(*info->samplers));
+   memcpy(saved->cso_samplers, info->cso_samplers,
+          sizeof(info->cso_samplers));
+   memcpy(saved->samplers, info->samplers, sizeof(info->samplers));
 }
 
 
@@ -1314,18 +1303,16 @@ cso_restore_fragment_samplers(struct cso_context *ctx)
 {
    struct sampler_info *info = &ctx->samplers[PIPE_SHADER_FRAGMENT];
    struct sampler_info *saved = &ctx->fragment_samplers_saved;
-   int delta = (int)info->nr_samplers - saved->nr_samplers;
 
    memcpy(info->cso_samplers, saved->cso_samplers,
-          saved->nr_samplers * sizeof(*info->cso_samplers));
-   memcpy(info->samplers, saved->samplers,
-          saved->nr_samplers * sizeof(*info->samplers));
-
-   if (delta > 0) {
-      memset(&info->cso_samplers[saved->nr_samplers], 0,
-             delta * sizeof(*info->cso_samplers));
-      memset(&info->samplers[saved->nr_samplers], 0,
-             delta * sizeof(*info->samplers));
+          sizeof(info->cso_samplers));
+   memcpy(info->samplers, saved->samplers, sizeof(info->samplers));
+
+   for (int i = PIPE_MAX_SAMPLERS - 1; i >= 0; i--) {
+      if (info->samplers[i]) {
+         ctx->max_sampler_seen = i;
+         break;
+      }
    }
 
    cso_single_sampler_done(ctx, PIPE_SHADER_FRAGMENT);