iris: Remove the CCS_D fallback
[mesa.git] / src / gallium / drivers / iris / iris_blorp.c
index 7995325239e178d8921e86eaccd7586b71737972..34f40f7fa614162bfdb4665d0336d5787757e54d 100644 (file)
@@ -4,30 +4,47 @@
  * 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
- * on the rights to use, copy, modify, merge, publish, distribute, sub
- * license, 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 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 above copyright notice and this permission notice 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 NON-INFRINGEMENT. IN NO EVENT SHALL
- * THE AUTHOR(S) AND/OR THEIR SUPPLIERS 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.
+ * 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 iris_blorp.c
+ *
+ * ============================= GENXML CODE =============================
+ *              [This file is compiled once per generation.]
+ * =======================================================================
+ *
+ * GenX specific code for working with BLORP (blitting, resolves, clears
+ * on the 3D engine).  This provides the driver-specific hooks needed to
+ * implement the BLORP API.
+ *
+ * See iris_blit.c, iris_clear.c, and so on.
+ */
+
 #include <assert.h>
 
 #include "iris_batch.h"
 #include "iris_resource.h"
 #include "iris_context.h"
 
-#include "blorp/blorp_genX_exec.h"
 #include "util/u_upload_mgr.h"
+#include "intel/common/gen_l3_config.h"
+
+#define BLORP_USE_SOFTPIN
+#include "blorp/blorp_genX_exec.h"
 
 static uint32_t *
 stream_state(struct iris_batch *batch,
@@ -42,10 +59,20 @@ stream_state(struct iris_batch *batch,
 
    u_upload_alloc(uploader, 0, size, alignment, out_offset, &res, &ptr);
 
-   *out_bo = iris_resource_bo(res);
-   iris_use_pinned_bo(batch, *out_bo, false);
+   struct iris_bo *bo = iris_resource_bo(res);
+   iris_use_pinned_bo(batch, bo, false, IRIS_DOMAIN_NONE);
+
+   iris_record_state_size(batch->state_sizes,
+                          bo->gtt_offset + *out_offset, size);
 
-   *out_offset += iris_bo_offset_from_base_address(*out_bo);
+   /* If the caller has asked for a BO, we leave them the responsibility of
+    * adding bo->gtt_offset (say, by handing an address to genxml).  If not,
+    * we assume they want the offset from a base address.
+    */
+   if (out_bo)
+      *out_bo = bo;
+   else
+      *out_offset += iris_bo_offset_from_base_address(bo);
 
    pipe_resource_reference(&res, NULL);
 
@@ -66,7 +93,8 @@ combine_and_pin_address(struct blorp_batch *blorp_batch,
    struct iris_batch *batch = blorp_batch->driver_batch;
    struct iris_bo *bo = addr.buffer;
 
-   iris_use_pinned_bo(batch, bo, addr.reloc_flags & RELOC_WRITE);
+   iris_use_pinned_bo(batch, bo, addr.reloc_flags & RELOC_WRITE,
+                      IRIS_DOMAIN_NONE);
 
    /* Assume this is a general address, not relative to a base. */
    return bo->gtt_offset + addr.offset;
@@ -96,7 +124,7 @@ blorp_get_surface_address(struct blorp_batch *blorp_batch,
 UNUSED static struct blorp_address
 blorp_get_surface_base_address(UNUSED struct blorp_batch *blorp_batch)
 {
-   return (struct blorp_address) { .offset = IRIS_MEMZONE_SURFACE_START };
+   return (struct blorp_address) { .offset = IRIS_MEMZONE_BINDER_START };
 }
 
 static void *
@@ -107,10 +135,9 @@ blorp_alloc_dynamic_state(struct blorp_batch *blorp_batch,
 {
    struct iris_context *ice = blorp_batch->blorp->driver_ctx;
    struct iris_batch *batch = blorp_batch->driver_batch;
-   struct iris_bo *bo;
 
    return stream_state(batch, ice->state.dynamic_uploader,
-                       size, alignment, offset, &bo);
+                       size, alignment, offset, NULL);
 }
 
 static void
@@ -123,19 +150,22 @@ blorp_alloc_binding_table(struct blorp_batch *blorp_batch,
                           void **surface_maps)
 {
    struct iris_context *ice = blorp_batch->blorp->driver_ctx;
+   struct iris_binder *binder = &ice->state.binder;
    struct iris_batch *batch = blorp_batch->driver_batch;
-   struct iris_bo *bo;
 
-   uint32_t *bt_map = iris_binder_reserve(&ice->state.binder,
-                                          num_entries * sizeof(uint32_t),
-                                          bt_offset);
+   *bt_offset = iris_binder_reserve(ice, num_entries * sizeof(uint32_t));
+   uint32_t *bt_map = binder->map + *bt_offset;
 
    for (unsigned i = 0; i < num_entries; i++) {
       surface_maps[i] = stream_state(batch, ice->state.surface_uploader,
                                      state_size, state_alignment,
-                                     &surface_offsets[i], &bo);
-      bt_map[i] = surface_offsets[i];
+                                     &surface_offsets[i], NULL);
+      bt_map[i] = surface_offsets[i] - (uint32_t) binder->bo->gtt_offset;
    }
+
+   iris_use_pinned_bo(batch, binder->bo, false, IRIS_DOMAIN_NONE);
+
+   batch->screen->vtbl.update_surface_base_address(batch, binder);
 }
 
 static void *
@@ -154,30 +184,30 @@ blorp_alloc_vertex_buffer(struct blorp_batch *blorp_batch,
    *addr = (struct blorp_address) {
       .buffer = bo,
       .offset = offset,
-      // XXX: Broadwell MOCS
-      .mocs = I915_MOCS_CACHED,
+      .mocs = iris_mocs(bo, &batch->screen->isl_dev),
    };
 
    return map;
 }
 
 /**
- * See vf_invalidate_for_vb_48b_transitions in iris_state.c.
- * XXX: actually add this
+ * See iris_upload_render_state's IRIS_DIRTY_VERTEX_BUFFERS handling for
+ * a comment about why these VF invalidations are needed.
  */
 static void
-blorp_vf_invalidate_for_vb_48b_transitions(struct blorp_batch *batch,
+blorp_vf_invalidate_for_vb_48b_transitions(struct blorp_batch *blorp_batch,
                                            const struct blorp_address *addrs,
+                                           UNUSED uint32_t *sizes,
                                            unsigned num_vbs)
 {
-#if 0
+#if GEN_GEN < 11
    struct iris_context *ice = blorp_batch->blorp->driver_ctx;
    struct iris_batch *batch = blorp_batch->driver_batch;
    bool need_invalidate = false;
 
    for (unsigned i = 0; i < num_vbs; i++) {
       struct iris_bo *bo = addrs[i].buffer;
-      uint16_t high_bits = bo ? bo->gtt_offset >> 32u : 0;
+      uint16_t high_bits = bo->gtt_offset >> 32u;
 
       if (high_bits != ice->state.last_vbo_high_bits[i]) {
          need_invalidate = true;
@@ -186,17 +216,23 @@ blorp_vf_invalidate_for_vb_48b_transitions(struct blorp_batch *batch,
    }
 
    if (need_invalidate) {
-      iris_emit_pipe_control_flush(batch, PIPE_CONTROL_VF_CACHE_INVALIDATE);
+      iris_emit_pipe_control_flush(batch,
+                                   "workaround: VF cache 32-bit key [blorp]",
+                                   PIPE_CONTROL_VF_CACHE_INVALIDATE |
+                                   PIPE_CONTROL_CS_STALL);
    }
 #endif
 }
 
 static struct blorp_address
-blorp_get_workaround_page(struct blorp_batch *blorp_batch)
+blorp_get_workaround_address(struct blorp_batch *blorp_batch)
 {
    struct iris_batch *batch = blorp_batch->driver_batch;
 
-   return (struct blorp_address) { .buffer = batch->screen->workaround_bo };
+   return (struct blorp_address) {
+      .buffer = batch->screen->workaround_address.bo,
+      .offset = batch->screen->workaround_address.offset,
+   };
 }
 
 static void
@@ -209,18 +245,11 @@ blorp_flush_range(UNUSED struct blorp_batch *blorp_batch,
     */
 }
 
-static void
-blorp_emit_urb_config(struct blorp_batch *blorp_batch,
-                      unsigned vs_entry_size,
-                      UNUSED unsigned sf_entry_size)
+static const struct gen_l3_config *
+blorp_get_l3_config(struct blorp_batch *blorp_batch)
 {
-   // XXX: URB...
-#if 0
-   if (ice->urb.vsize >= vs_entry_size)
-      return;
-
-   gen7_upload_urb(ice, vs_entry_size, false, false);
-#endif
+   struct iris_batch *batch = blorp_batch->driver_batch;
+   return batch->screen->l3_config_3d;
 }
 
 static void
@@ -230,90 +259,119 @@ iris_blorp_exec(struct blorp_batch *blorp_batch,
    struct iris_context *ice = blorp_batch->blorp->driver_ctx;
    struct iris_batch *batch = blorp_batch->driver_batch;
 
-   /* Flush the sampler and render caches.  We definitely need to flush the
-    * sampler cache so that we get updated contents from the render cache for
-    * the glBlitFramebuffer() source.  Also, we are sometimes warned in the
-    * docs to flush the cache between reinterpretations of the same surface
-    * data with different formats, which blorp does for stencil and depth
-    * data.
+#if GEN_GEN >= 11
+   /* The PIPE_CONTROL command description says:
+    *
+    *    "Whenever a Binding Table Index (BTI) used by a Render Target Message
+    *     points to a different RENDER_SURFACE_STATE, SW must issue a Render
+    *     Target Cache Flush by enabling this bit. When render target flush
+    *     is set due to new association of BTI, PS Scoreboard Stall bit must
+    *     be set in this packet."
+    */
+   iris_emit_pipe_control_flush(batch,
+                                "workaround: RT BTI change [blorp]",
+                                PIPE_CONTROL_RENDER_TARGET_FLUSH |
+                                PIPE_CONTROL_STALL_AT_SCOREBOARD);
+#endif
+
+   /* Flush the render cache in cases where the same surface is reinterpreted
+    * with a differernt format, which blorp does for stencil and depth data
+    * among other things.  Invalidation of sampler caches and flushing of any
+    * caches which had previously written the source surfaces should already
+    * have been handled by the caller.
     */
-   if (params->src.enabled)
-      iris_cache_flush_for_read(batch, params->src.addr.buffer);
    if (params->dst.enabled) {
       iris_cache_flush_for_render(batch, params->dst.addr.buffer,
                                   params->dst.view.format,
                                   params->dst.aux_usage);
    }
-   if (params->depth.enabled)
-      iris_cache_flush_for_depth(batch, params->depth.addr.buffer);
-   if (params->stencil.enabled)
-      iris_cache_flush_for_depth(batch, params->stencil.addr.buffer);
 
    iris_require_command_space(batch, 1400);
-   //iris_require_statebuffer_space(ice, 600); // XXX: THIS.  Need this.
-   batch->no_wrap = true;
-
-   // XXX: Emit L3 state
 
 #if GEN_GEN == 8
-   // XXX: PMA - gen8_write_pma_stall_bits(ice, 0);
+   genX(update_pma_fix)(ice, batch, false);
 #endif
 
-   // XXX: knock this off...land Jason's i965 patches...
-   blorp_emit(blorp_batch, GENX(3DSTATE_DRAWING_RECTANGLE), rect) {
-      rect.ClippedDrawingRectangleXMax = MAX2(params->x1, params->x0) - 1;
-      rect.ClippedDrawingRectangleYMax = MAX2(params->y1, params->y0) - 1;
+   const unsigned scale = params->fast_clear_op ? UINT_MAX : 1;
+   if (ice->state.current_hash_scale != scale) {
+      genX(emit_hashing_mode)(ice, batch, params->x1 - params->x0,
+                              params->y1 - params->y0, scale);
    }
 
-   blorp_exec(blorp_batch, params);
+#if GEN_GEN >= 12
+   genX(invalidate_aux_map_state)(batch);
+#endif
 
-   batch->no_wrap = false;
+   iris_handle_always_flush_cache(batch);
 
-   // XXX: aperture checks?
+   blorp_exec(blorp_batch, params);
+
+   iris_handle_always_flush_cache(batch);
 
    /* We've smashed all state compared to what the normal 3D pipeline
     * rendering tracks for GL.
     */
-   // XXX: skip some if (!(batch->flags & BLORP_BATCH_NO_EMIT_DEPTH_STENCIL))
-   ice->state.dirty |= ~(IRIS_DIRTY_POLYGON_STIPPLE |
-                         IRIS_DIRTY_LINE_STIPPLE);
-
-#if 0
-   ice->state.dirty |= IRIS_DIRTY_VERTEX_BUFFERS |
-                       IRIS_DIRTY_COLOR_CALC_STATE |
-                       IRIS_DIRTY_CONSTANTS_VS |
-                       IRIS_DIRTY_CONSTANTS_TCS |
-                       IRIS_DIRTY_CONSTANTS_TES |
-                       IRIS_DIRTY_CONSTANTS_GS |
-                       IRIS_DIRTY_CONSTANTS_PS |
-                       IRIS_DIRTY_CONSTANTS_PS |
-                       IRIS_DIRTY_SAMPLER_STATES_VS |
-                       IRIS_DIRTY_SAMPLER_STATES_TCS |
-                       IRIS_DIRTY_SAMPLER_STATES_TES |
-                       IRIS_DIRTY_SAMPLER_STATES_GS |
-                       IRIS_DIRTY_SAMPLER_STATES_PS |
-                       IRIS_DIRTY_SAMPLER_STATES_PS |
-                       IRIS_DIRTY_MULTISAMPLE |
-                       IRIS_DIRTY_SAMPLE_MASK |
-                       IRIS_DIRTY_VS |
-                       IRIS_DIRTY_TCS |
-                       IRIS_DIRTY_TES |
-                       // IRIS_DIRTY_STREAMOUT |
-                       IRIS_DIRTY_GS |
-                       IRIS_DIRTY_CLIP |
-                       IRIS_DIRTY_FS |
-                       IRIS_DIRTY_CC_VIEWPORT |
-#endif
 
-   if (params->dst.enabled) {
-      iris_render_cache_add_bo(batch, params->dst.addr.buffer,
-                               params->dst.view.format,
-                               params->dst.aux_usage);
+   uint64_t skip_bits = (IRIS_DIRTY_POLYGON_STIPPLE |
+                         IRIS_DIRTY_SO_BUFFERS |
+                         IRIS_DIRTY_SO_DECL_LIST |
+                         IRIS_DIRTY_LINE_STIPPLE |
+                         IRIS_ALL_DIRTY_FOR_COMPUTE |
+                         IRIS_DIRTY_SCISSOR_RECT |
+                         IRIS_DIRTY_VF |
+                         IRIS_DIRTY_SF_CL_VIEWPORT);
+   uint64_t skip_stage_bits = (IRIS_ALL_STAGE_DIRTY_FOR_COMPUTE |
+                               IRIS_STAGE_DIRTY_UNCOMPILED_VS |
+                               IRIS_STAGE_DIRTY_UNCOMPILED_TCS |
+                               IRIS_STAGE_DIRTY_UNCOMPILED_TES |
+                               IRIS_STAGE_DIRTY_UNCOMPILED_GS |
+                               IRIS_STAGE_DIRTY_UNCOMPILED_FS |
+                               IRIS_STAGE_DIRTY_SAMPLER_STATES_VS |
+                               IRIS_STAGE_DIRTY_SAMPLER_STATES_TCS |
+                               IRIS_STAGE_DIRTY_SAMPLER_STATES_TES |
+                               IRIS_STAGE_DIRTY_SAMPLER_STATES_GS);
+
+   if (!ice->shaders.uncompiled[MESA_SHADER_TESS_EVAL]) {
+      /* BLORP disabled tessellation, that's fine for the next draw */
+      skip_stage_bits |= IRIS_STAGE_DIRTY_TCS |
+                         IRIS_STAGE_DIRTY_TES |
+                         IRIS_STAGE_DIRTY_CONSTANTS_TCS |
+                         IRIS_STAGE_DIRTY_CONSTANTS_TES |
+                         IRIS_STAGE_DIRTY_BINDINGS_TCS |
+                         IRIS_STAGE_DIRTY_BINDINGS_TES;
+   }
+
+   if (!ice->shaders.uncompiled[MESA_SHADER_GEOMETRY]) {
+      /* BLORP disabled geometry shaders, that's fine for the next draw */
+      skip_stage_bits |= IRIS_STAGE_DIRTY_GS |
+                         IRIS_STAGE_DIRTY_CONSTANTS_GS |
+                         IRIS_STAGE_DIRTY_BINDINGS_GS;
    }
+
+   /* we can skip flagging IRIS_DIRTY_DEPTH_BUFFER, if
+    * BLORP_BATCH_NO_EMIT_DEPTH_STENCIL is set.
+    */
+   if (blorp_batch->flags & BLORP_BATCH_NO_EMIT_DEPTH_STENCIL)
+      skip_bits |= IRIS_DIRTY_DEPTH_BUFFER;
+
+   if (!params->wm_prog_data)
+      skip_bits |= IRIS_DIRTY_BLEND_STATE | IRIS_DIRTY_PS_BLEND;
+
+   ice->state.dirty |= ~skip_bits;
+   ice->state.stage_dirty |= ~skip_stage_bits;
+
+   if (params->src.enabled)
+      iris_bo_bump_seqno(params->src.addr.buffer, batch->next_seqno,
+                         IRIS_DOMAIN_OTHER_READ);
+   if (params->dst.enabled)
+      iris_bo_bump_seqno(params->dst.addr.buffer, batch->next_seqno,
+                         IRIS_DOMAIN_RENDER_WRITE);
    if (params->depth.enabled)
-      iris_depth_cache_add_bo(batch, params->depth.addr.buffer);
+      iris_bo_bump_seqno(params->depth.addr.buffer, batch->next_seqno,
+                         IRIS_DOMAIN_DEPTH_WRITE);
    if (params->stencil.enabled)
-      iris_depth_cache_add_bo(batch, params->stencil.addr.buffer);
+      iris_bo_bump_seqno(params->stencil.addr.buffer, batch->next_seqno,
+                         IRIS_DOMAIN_DEPTH_WRITE);
 }
 
 void
@@ -323,6 +381,7 @@ genX(init_blorp)(struct iris_context *ice)
 
    blorp_init(&ice->blorp, ice, &screen->isl_dev);
    ice->blorp.compiler = screen->compiler;
-
-   ice->vtbl.blorp_exec = iris_blorp_exec;
+   ice->blorp.lookup_shader = iris_blorp_lookup_shader;
+   ice->blorp.upload_shader = iris_blorp_upload_shader;
+   ice->blorp.exec = iris_blorp_exec;
 }