i965/gen4: Move VS state to state streaming.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>

diff --git a/src/mesa/drivers/dri/i965/brw_context.h b/src/mesa/drivers/dri/i965/brw_context.h
index 0876c3e94e62803eaa53da0a45b874e3e0be4bd5..a81d6157390dbaf8121ae63a2da10f60b961d440 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_context.h
+++ b/src/mesa/drivers/dri/i965/brw_context.h
@@ -631,8 +631,8 @@ struct brw_context
       int8_t *constant_map; /* variable array following prog_data */
 
       drm_intel_bo *prog_bo;
-      drm_intel_bo *state_bo;
       drm_intel_bo *const_bo;
+      uint32_t state_offset;
 
       /** Binding table of pointers to surf_bo entries */
       uint32_t bind_bo_offset;

diff --git a/src/mesa/drivers/dri/i965/brw_misc_state.c b/src/mesa/drivers/dri/i965/brw_misc_state.c
index 7325a2b6c8dab55d34c970d4dcb6f8b11cfadbec..b61a7ad099ef4a89d97c58aabc21bfc7836da4de 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_misc_state.c
+++ b/src/mesa/drivers/dri/i965/brw_misc_state.c
@@ -143,7 +143,8 @@ static void upload_pipelined_state_pointers(struct brw_context *brw )
 
    BEGIN_BATCH(7);
    OUT_BATCH(_3DSTATE_PIPELINED_POINTERS << 16 | (7 - 2));
-   OUT_RELOC(brw->vs.state_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 0);
+   OUT_RELOC(intel->batch.bo, I915_GEM_DOMAIN_INSTRUCTION, 0,
+            brw->vs.state_offset);
    if (brw->gs.prog_active)
       OUT_RELOC(brw->gs.state_bo, I915_GEM_DOMAIN_INSTRUCTION, 0, 1);
    else
@@ -163,7 +164,6 @@ static void upload_pipelined_state_pointers(struct brw_context *brw )
 
 static void prepare_psp_urb_cbs(struct brw_context *brw)
 {
-   brw_add_validated_bo(brw, brw->vs.state_bo);
    brw_add_validated_bo(brw, brw->gs.state_bo);
    brw_add_validated_bo(brw, brw->clip.state_bo);
 }

diff --git a/src/mesa/drivers/dri/i965/brw_state_dump.c b/src/mesa/drivers/dri/i965/brw_state_dump.c
index 930eb5155dbd02500e3c101992cfdb04d8dc0ecd..4313b2aaa9896b913e0bb589fca5dcef46ae0787 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_state_dump.c
+++ b/src/mesa/drivers/dri/i965/brw_state_dump.c
@@ -388,7 +388,8 @@ void brw_debug_batch(struct intel_context *intel)
    dump_wm_sampler_state(brw);
 
    if (intel->gen < 6)
-       state_struct_out("VS", brw->vs.state_bo, 0, sizeof(struct brw_vs_unit_state));
+       state_struct_out("VS", intel->batch.bo, brw->vs.state_offset,
+                       sizeof(struct brw_vs_unit_state));
    brw_debug_prog("VS prog", brw->vs.prog_bo);
 
    if (intel->gen < 6)

diff --git a/src/mesa/drivers/dri/i965/brw_vs_state.c b/src/mesa/drivers/dri/i965/brw_vs_state.c
index c3a7cc247c560827b6ac8acac0f249353846cb79..1eee5b7e5de8e593f88149a7df826c8859215a72 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_vs_state.c
+++ b/src/mesa/drivers/dri/i965/brw_vs_state.c
@@ -49,48 +49,19 @@ struct brw_vs_unit_key {
 };
 
 static void
-vs_unit_populate_key(struct brw_context *brw, struct brw_vs_unit_key *key)
-{
-   struct gl_context *ctx = &brw->intel.ctx;
-
-   memset(key, 0, sizeof(*key));
-
-   /* CACHE_NEW_VS_PROG */
-   key->total_grf = brw->vs.prog_data->total_grf;
-   key->urb_entry_read_length = brw->vs.prog_data->urb_read_length;
-   key->curb_entry_read_length = brw->vs.prog_data->curb_read_length;
-
-   /* BRW_NEW_URB_FENCE */
-   key->nr_urb_entries = brw->urb.nr_vs_entries;
-   key->urb_size = brw->urb.vsize;
-
-   /* BRW_NEW_NR_VS_SURFACES */
-   key->nr_surfaces = brw->vs.nr_surfaces;
-
-   /* BRW_NEW_CURBE_OFFSETS, _NEW_TRANSFORM */
-   if (ctx->Transform.ClipPlanesEnabled) {
-      /* Note that we read in the userclip planes as well, hence
-       * clip_start:
-       */
-      key->curbe_offset = brw->curbe.clip_start;
-   }
-   else {
-      key->curbe_offset = brw->curbe.vs_start;
-   }
-}
-
-static drm_intel_bo *
-vs_unit_create_from_key(struct brw_context *brw, struct brw_vs_unit_key *key)
+brw_prepare_vs_unit(struct brw_context *brw)
 {
    struct intel_context *intel = &brw->intel;
-   struct brw_vs_unit_state vs;
-   drm_intel_bo *bo;
+   struct gl_context *ctx = &intel->ctx;
+   struct brw_vs_unit_state *vs;
 
-   memset(&vs, 0, sizeof(vs));
+   vs = brw_state_batch(brw, sizeof(*vs), 32, &brw->vs.state_offset);
+   memset(vs, 0, sizeof(*vs));
 
-   vs.thread0.kernel_start_pointer = brw->vs.prog_bo->offset >> 6; /* reloc */
-   vs.thread0.grf_reg_count = ALIGN(key->total_grf, 16) / 16 - 1;
-   vs.thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754;
+   /* CACHE_NEW_VS_PROG */
+   vs->thread0.kernel_start_pointer = brw->vs.prog_bo->offset >> 6; /* reloc */
+   vs->thread0.grf_reg_count = ALIGN(brw->vs.prog_data->total_grf, 16) / 16 - 1;
+   vs->thread1.floating_point_mode = BRW_FLOATING_POINT_NON_IEEE_754;
    /* Choosing multiple program flow means that we may get 2-vertex threads,
     * which will have the channel mask for dwords 4-7 enabled in the thread,
     * and those dwords will be written to the second URB handle when we
@@ -103,21 +74,34 @@ vs_unit_create_from_key(struct brw_context *brw, struct brw_vs_unit_key *key)
     * The most notable and reliably failing application is the Humus
     * demo "CelShading"
    */
-   vs.thread1.single_program_flow = (intel->gen == 5);
+   vs->thread1.single_program_flow = (intel->gen == 5);
 
+   /* BRW_NEW_NR_VS_SURFACES */
    if (intel->gen == 5)
-      vs.thread1.binding_table_entry_count = 0; /* hardware requirement */
+      vs->thread1.binding_table_entry_count = 0; /* hardware requirement */
    else
-      vs.thread1.binding_table_entry_count = key->nr_surfaces;
+      vs->thread1.binding_table_entry_count = brw->vs.nr_surfaces;
+
+   vs->thread3.urb_entry_read_length = brw->vs.prog_data->urb_read_length;
+   vs->thread3.const_urb_entry_read_length = brw->vs.prog_data->curb_read_length;
+   vs->thread3.dispatch_grf_start_reg = 1;
+   vs->thread3.urb_entry_read_offset = 0;
 
-   vs.thread3.urb_entry_read_length = key->urb_entry_read_length;
-   vs.thread3.const_urb_entry_read_length = key->curb_entry_read_length;
-   vs.thread3.dispatch_grf_start_reg = 1;
-   vs.thread3.urb_entry_read_offset = 0;
-   vs.thread3.const_urb_entry_read_offset = key->curbe_offset * 2;
+   /* BRW_NEW_CURBE_OFFSETS, _NEW_TRANSFORM */
+   if (ctx->Transform.ClipPlanesEnabled) {
+      /* Note that we read in the userclip planes as well, hence
+       * clip_start:
+       */
+      vs->thread3.const_urb_entry_read_offset = brw->curbe.clip_start * 2;
+   }
+   else {
+      vs->thread3.const_urb_entry_read_offset = brw->curbe.vs_start * 2;
+   }
 
+
+   /* BRW_NEW_URB_FENCE */
    if (intel->gen == 5) {
-      switch (key->nr_urb_entries) {
+      switch (brw->urb.nr_vs_entries) {
       case 8:
       case 12:
       case 16:
@@ -129,13 +113,13 @@ vs_unit_create_from_key(struct brw_context *brw, struct brw_vs_unit_key *key)
       case 192:
       case 224:
       case 256:
-        vs.thread4.nr_urb_entries = key->nr_urb_entries >> 2;
+        vs->thread4.nr_urb_entries = brw->urb.nr_vs_entries >> 2;
         break;
       default:
         assert(0);
       }
    } else {
-      switch (key->nr_urb_entries) {
+      switch (brw->urb.nr_vs_entries) {
       case 8:
       case 12:
       case 16:
@@ -147,63 +131,45 @@ vs_unit_create_from_key(struct brw_context *brw, struct brw_vs_unit_key *key)
       default:
         assert(0);
       }
-      vs.thread4.nr_urb_entries = key->nr_urb_entries;
+      vs->thread4.nr_urb_entries = brw->urb.nr_vs_entries;
    }
 
-   vs.thread4.urb_entry_allocation_size = key->urb_size - 1;
+   vs->thread4.urb_entry_allocation_size = brw->urb.vsize - 1;
 
-   vs.thread4.max_threads = CLAMP(key->nr_urb_entries / 2,
-                                 1, brw->vs_max_threads) - 1;
+   vs->thread4.max_threads = CLAMP(brw->urb.nr_vs_entries / 2,
+                                  1, brw->vs_max_threads) - 1;
 
    /* No samplers for ARB_vp programs:
     */
    /* It has to be set to 0 for Ironlake
     */
-   vs.vs5.sampler_count = 0;
+   vs->vs5.sampler_count = 0;
 
    if (unlikely(INTEL_DEBUG & DEBUG_STATS))
-      vs.thread4.stats_enable = 1;
+      vs->thread4.stats_enable = 1;
 
    /* Vertex program always enabled:
     */
-   vs.vs6.vs_enable = 1;
-
-   bo = brw_upload_cache(&brw->cache, BRW_VS_UNIT,
-                        key, sizeof(*key),
-                        &brw->vs.prog_bo, 1,
-                        &vs, sizeof(vs));
+   vs->vs6.vs_enable = 1;
 
    /* Emit VS program relocation */
-   drm_intel_bo_emit_reloc(bo, offsetof(struct brw_vs_unit_state, thread0),
-                          brw->vs.prog_bo, vs.thread0.grf_reg_count << 1,
+   drm_intel_bo_emit_reloc(intel->batch.bo, (brw->vs.state_offset +
+                                            offsetof(struct brw_vs_unit_state,
+                                                     thread0)),
+                          brw->vs.prog_bo, vs->thread0.grf_reg_count << 1,
                           I915_GEM_DOMAIN_INSTRUCTION, 0);
 
-   return bo;
-}
-
-static void prepare_vs_unit(struct brw_context *brw)
-{
-   struct brw_vs_unit_key key;
-
-   vs_unit_populate_key(brw, &key);
-
-   drm_intel_bo_unreference(brw->vs.state_bo);
-   brw->vs.state_bo = brw_search_cache(&brw->cache, BRW_VS_UNIT,
-                                      &key, sizeof(key),
-                                      &brw->vs.prog_bo, 1,
-                                      NULL);
-   if (brw->vs.state_bo == NULL) {
-      brw->vs.state_bo = vs_unit_create_from_key(brw, &key);
-   }
+   brw->state.dirty.cache |= CACHE_NEW_VS_UNIT;
 }
 
 const struct brw_tracked_state brw_vs_unit = {
    .dirty = {
       .mesa  = _NEW_TRANSFORM,
-      .brw   = (BRW_NEW_CURBE_OFFSETS |
+      .brw   = (BRW_NEW_BATCH |
+               BRW_NEW_CURBE_OFFSETS |
                 BRW_NEW_NR_VS_SURFACES |
                BRW_NEW_URB_FENCE),
       .cache = CACHE_NEW_VS_PROG
    },
-   .prepare = prepare_vs_unit,
+   .prepare = brw_prepare_vs_unit,
 };

diff --git a/src/mesa/drivers/dri/i965/brw_vtbl.c b/src/mesa/drivers/dri/i965/brw_vtbl.c
index b1b2e42d3deb2230adaac15976cc07c69f426fe2..8d1497c8215e85bfb822eb7f1eaf7aa3653f4bc7 100644 (file)
--- a/src/mesa/drivers/dri/i965/brw_vtbl.c
+++ b/src/mesa/drivers/dri/i965/brw_vtbl.c
@@ -60,7 +60,6 @@ dri_bo_release(drm_intel_bo **bo)
 static void brw_destroy_context( struct intel_context *intel )
 {
    struct brw_context *brw = brw_context(&intel->ctx);
-   int i;
 
    brw_destroy_state(brw);
    brw_draw_destroy( brw );
@@ -77,7 +76,6 @@ static void brw_destroy_context( struct intel_context *intel )
 
    dri_bo_release(&brw->curbe.curbe_bo);
    dri_bo_release(&brw->vs.prog_bo);
-   dri_bo_release(&brw->vs.state_bo);
    dri_bo_release(&brw->vs.const_bo);
    dri_bo_release(&brw->gs.prog_bo);
    dri_bo_release(&brw->gs.state_bo);