assert(batch->blorp->driver_ctx == batch->driver_batch);
struct brw_context *brw = batch->driver_batch;
- intel_batchbuffer_begin(brw, n, RENDER_RING);
+ intel_batchbuffer_begin(brw, n);
uint32_t *map = brw->batch.map_next;
brw->batch.map_next += n;
intel_batchbuffer_advance(brw);
uint32_t offset;
if (GEN_GEN < 6 && brw_ptr_in_state_buffer(&brw->batch, location)) {
- offset = (char *)location - (char *)brw->batch.state_map;
+ offset = (char *)location - (char *)brw->batch.state.map;
return brw_state_reloc(&brw->batch, offset,
address.buffer, address.offset + delta,
address.reloc_flags);
assert(!brw_ptr_in_state_buffer(&brw->batch, location));
- offset = (char *)location - (char *)brw->batch.map;
+ offset = (char *)location - (char *)brw->batch.batch.map;
return brw_batch_reloc(&brw->batch, offset,
address.buffer, address.offset + delta,
address.reloc_flags);
brw_state_reloc(&brw->batch, ss_offset, bo, address.offset + delta,
address.reloc_flags);
- void *reloc_ptr = (void *)brw->batch.state_map + ss_offset;
+ void *reloc_ptr = (void *)brw->batch.state.map + ss_offset;
#if GEN_GEN >= 8
*(uint64_t *)reloc_ptr = reloc_val;
#else
#endif
}
+static uint64_t
+blorp_get_surface_address(struct blorp_batch *blorp_batch,
+ struct blorp_address address)
+{
+ /* We'll let blorp_surface_reloc write the address. */
+ return 0ull;
+}
+
+#if GEN_GEN >= 7 && GEN_GEN < 10
+static struct blorp_address
+blorp_get_surface_base_address(struct blorp_batch *batch)
+{
+ assert(batch->blorp->driver_ctx == batch->driver_batch);
+ struct brw_context *brw = batch->driver_batch;
+ return (struct blorp_address) {
+ .buffer = brw->batch.state.bo,
+ .offset = 0,
+ };
+}
+#endif
+
static void *
blorp_alloc_dynamic_state(struct blorp_batch *batch,
uint32_t size,
void *data = brw_state_batch(brw, size, 64, &offset);
*addr = (struct blorp_address) {
- .buffer = brw->batch.state_bo,
+ .buffer = brw->batch.state.bo,
.offset = offset,
+
+ /* The VF cache designers apparently cut corners, and made the cache
+ * only consider the bottom 32 bits of memory addresses. If you happen
+ * to have two vertex buffers which get placed exactly 4 GiB apart and
+ * use them in back-to-back draw calls, you can get collisions. To work
+ * around this problem, we restrict vertex buffers to the low 32 bits of
+ * the address space.
+ */
+ .reloc_flags = RELOC_32BIT,
+
+#if GEN_GEN == 11
+ .mocs = ICL_MOCS_WB,
+#elif GEN_GEN == 10
+ .mocs = CNL_MOCS_WB,
+#elif GEN_GEN == 9
+ .mocs = SKL_MOCS_WB,
+#elif GEN_GEN == 8
+ .mocs = BDW_MOCS_WB,
+#elif GEN_GEN == 7
+ .mocs = GEN7_MOCS_L3,
+#elif GEN_GEN > 6
+#error "Missing MOCS setting!"
+#endif
};
return data;
}
+/**
+ * See vf_invalidate_for_vb_48b_transitions in genX_state_upload.c.
+ */
+static void
+blorp_vf_invalidate_for_vb_48b_transitions(struct blorp_batch *batch,
+ const struct blorp_address *addrs,
+ unsigned num_vbs)
+{
+#if GEN_GEN >= 8 && GEN_GEN < 11
+ struct brw_context *brw = batch->driver_batch;
+ bool need_invalidate = false;
+
+ for (unsigned i = 0; i < num_vbs; i++) {
+ struct brw_bo *bo = addrs[i].buffer;
+ uint16_t high_bits =
+ bo && (bo->kflags & EXEC_OBJECT_PINNED) ? bo->gtt_offset >> 32u : 0;
+
+ if (high_bits != brw->vb.last_bo_high_bits[i]) {
+ need_invalidate = true;
+ brw->vb.last_bo_high_bits[i] = high_bits;
+ }
+ }
+
+ if (need_invalidate) {
+ brw_emit_pipe_control_flush(brw, PIPE_CONTROL_VF_CACHE_INVALIDATE | PIPE_CONTROL_CS_STALL);
+ }
+#endif
+}
+
#if GEN_GEN >= 8
static struct blorp_address
blorp_get_workaround_page(struct blorp_batch *batch)
#endif
static void
-blorp_flush_range(struct blorp_batch *batch, void *start, size_t size)
+blorp_flush_range(UNUSED struct blorp_batch *batch, UNUSED void *start,
+ UNUSED size_t size)
{
/* All allocated states come from the batch which we will flush before we
* submit it. There's nothing for us to do here.
static void
blorp_emit_urb_config(struct blorp_batch *batch,
- unsigned vs_entry_size, unsigned sf_entry_size)
+ unsigned vs_entry_size,
+ UNUSED unsigned sf_entry_size)
{
assert(batch->blorp->driver_ctx == batch->driver_batch);
struct brw_context *brw = batch->driver_batch;
assert(batch->blorp->driver_ctx == batch->driver_batch);
struct brw_context *brw = batch->driver_batch;
struct gl_context *ctx = &brw->ctx;
- const uint32_t estimated_max_batch_usage = GEN_GEN >= 8 ? 1920 : 1700;
bool check_aperture_failed_once = false;
+#if GEN_GEN >= 11
+ /* The PIPE_CONTROL command description says:
+ *
+ * "Whenever a Binding Table Index (BTI) used by a Render Taget 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."
+ */
+ brw_emit_pipe_control_flush(brw,
+ PIPE_CONTROL_RENDER_TARGET_FLUSH |
+ PIPE_CONTROL_STALL_AT_SCOREBOARD);
+#endif
+
/* 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
* data.
*/
if (params->src.enabled)
- brw_render_cache_set_check_flush(brw, params->src.addr.buffer);
- brw_render_cache_set_check_flush(brw, params->dst.addr.buffer);
+ brw_cache_flush_for_read(brw, params->src.addr.buffer);
+ if (params->dst.enabled) {
+ brw_cache_flush_for_render(brw, params->dst.addr.buffer,
+ params->dst.view.format,
+ params->dst.aux_usage);
+ }
+ if (params->depth.enabled)
+ brw_cache_flush_for_depth(brw, params->depth.addr.buffer);
+ if (params->stencil.enabled)
+ brw_cache_flush_for_depth(brw, params->stencil.addr.buffer);
brw_select_pipeline(brw, BRW_RENDER_PIPELINE);
retry:
- intel_batchbuffer_require_space(brw, estimated_max_batch_usage, RENDER_RING);
+ intel_batchbuffer_require_space(brw, 1400);
+ brw_require_statebuffer_space(brw, 600);
intel_batchbuffer_save_state(brw);
- brw->no_batch_wrap = true;
+ check_aperture_failed_once |= intel_batchbuffer_saved_state_is_empty(brw);
+ brw->batch.no_wrap = true;
#if GEN_GEN == 6
/* Emit workaround flushes when we switch from drawing to blorping. */
gen8_write_pma_stall_bits(brw, 0);
#endif
+ const unsigned scale = params->fast_clear_op ? UINT_MAX : 1;
+ if (brw->current_hash_scale != scale) {
+ brw_emit_hashing_mode(brw, params->x1 - params->x0,
+ params->y1 - params->y0, scale);
+ }
+
blorp_emit(batch, GENX(3DSTATE_DRAWING_RECTANGLE), rect) {
rect.ClippedDrawingRectangleXMax = MAX2(params->x1, params->x0) - 1;
rect.ClippedDrawingRectangleYMax = MAX2(params->y1, params->y0) - 1;
blorp_exec(batch, params);
- brw->no_batch_wrap = false;
+ brw->batch.no_wrap = false;
/* Check if the blorp op we just did would make our batch likely to fail to
* map all the BOs into the GPU at batch exec time later. If so, flush the
!params->stencil.enabled;
brw->ib.index_size = -1;
- if (params->dst.enabled)
- brw_render_cache_set_add_bo(brw, params->dst.addr.buffer);
+ if (params->dst.enabled) {
+ brw_render_cache_add_bo(brw, params->dst.addr.buffer,
+ params->dst.view.format,
+ params->dst.aux_usage);
+ }
if (params->depth.enabled)
- brw_render_cache_set_add_bo(brw, params->depth.addr.buffer);
+ brw_depth_cache_add_bo(brw, params->depth.addr.buffer);
if (params->stencil.enabled)
- brw_render_cache_set_add_bo(brw, params->stencil.addr.buffer);
+ brw_depth_cache_add_bo(brw, params->stencil.addr.buffer);
}