static void
upload_3dstate_so_buffers(struct brw_context *brw)
{
- struct intel_context *intel = &brw->intel;
- struct gl_context *ctx = &intel->ctx;
- /* BRW_NEW_VERTEX_PROGRAM */
- const struct gl_shader_program *vs_prog =
- ctx->Shader.CurrentVertexProgram;
- const struct gl_transform_feedback_info *linked_xfb_info =
- &vs_prog->LinkedTransformFeedback;
+ struct gl_context *ctx = &brw->ctx;
/* BRW_NEW_TRANSFORM_FEEDBACK */
struct gl_transform_feedback_object *xfb_obj =
ctx->TransformFeedback.CurrentObject;
+ const struct gl_transform_feedback_info *linked_xfb_info =
+ &xfb_obj->shader_program->LinkedTransformFeedback;
int i;
/* Set up the up to 4 output buffers. These are the ranges defined in the
continue;
}
- bo = intel_bufferobj_buffer(intel, bufferobj, INTEL_WRITE_PART);
stride = linked_xfb_info->BufferStride[i] * 4;
start = xfb_obj->Offset[i];
assert(start % 4 == 0);
end = ALIGN(start + xfb_obj->Size[i], 4);
+ bo = intel_bufferobj_buffer(brw, bufferobj, start, end - start);
assert(end <= bo->size);
- /* If we don't have hardware contexts, then we reset our offsets at the
- * start of every batch, so we track the number of vertices written in
- * software and increment our pointers by that many.
- */
- if (!intel->hw_ctx) {
- start += brw->sol.offset_0_batch_start * stride;
- assert(start <= end);
- }
-
BEGIN_BATCH(4);
OUT_BATCH(_3DSTATE_SO_BUFFER << 16 | (4 - 2));
OUT_BATCH((i << SO_BUFFER_INDEX_SHIFT) | stride);
* stream. We only have one stream of rendering coming out of the GS unit, so
* we only emit stream 0 (low 16 bits) SO_DECLs.
*/
-static void
-upload_3dstate_so_decl_list(struct brw_context *brw,
- const struct brw_vue_map *vue_map)
+void
+gen7_upload_3dstate_so_decl_list(struct brw_context *brw,
+ const struct brw_vue_map *vue_map)
{
- struct intel_context *intel = &brw->intel;
- struct gl_context *ctx = &intel->ctx;
- /* BRW_NEW_VERTEX_PROGRAM */
- const struct gl_shader_program *vs_prog =
- ctx->Shader.CurrentVertexProgram;
+ struct gl_context *ctx = &brw->ctx;
/* BRW_NEW_TRANSFORM_FEEDBACK */
+ struct gl_transform_feedback_object *xfb_obj =
+ ctx->TransformFeedback.CurrentObject;
const struct gl_transform_feedback_info *linked_xfb_info =
- &vs_prog->LinkedTransformFeedback;
- int i;
- uint16_t so_decl[128];
- int buffer_mask = 0;
- int next_offset[4] = {0, 0, 0, 0};
+ &xfb_obj->shader_program->LinkedTransformFeedback;
+ uint16_t so_decl[MAX_VERTEX_STREAMS][128];
+ int buffer_mask[MAX_VERTEX_STREAMS] = {0, 0, 0, 0};
+ int next_offset[MAX_VERTEX_STREAMS] = {0, 0, 0, 0};
+ int decls[MAX_VERTEX_STREAMS] = {0, 0, 0, 0};
+ int max_decls = 0;
+ STATIC_ASSERT(ARRAY_SIZE(so_decl[0]) >= MAX_PROGRAM_OUTPUTS);
- STATIC_ASSERT(ARRAY_SIZE(so_decl) >= MAX_PROGRAM_OUTPUTS);
+ memset(so_decl, 0, sizeof(so_decl));
/* Construct the list of SO_DECLs to be emitted. The formatting of the
* command is feels strange -- each dword pair contains a SO_DECL per stream.
*/
- for (i = 0; i < linked_xfb_info->NumOutputs; i++) {
+ for (unsigned i = 0; i < linked_xfb_info->NumOutputs; i++) {
int buffer = linked_xfb_info->Outputs[i].OutputBuffer;
uint16_t decl = 0;
int varying = linked_xfb_info->Outputs[i].OutputRegister;
- unsigned component_mask =
- (1 << linked_xfb_info->Outputs[i].NumComponents) - 1;
+ const unsigned components = linked_xfb_info->Outputs[i].NumComponents;
+ unsigned component_mask = (1 << components) - 1;
+ unsigned stream_id = linked_xfb_info->Outputs[i].StreamId;
+
+ assert(stream_id < MAX_VERTEX_STREAMS);
- /* gl_PointSize is stored in VARYING_SLOT_PSIZ.w. */
+ /* gl_PointSize is stored in VARYING_SLOT_PSIZ.w
+ * gl_Layer is stored in VARYING_SLOT_PSIZ.y
+ * gl_ViewportIndex is stored in VARYING_SLOT_PSIZ.z
+ */
if (varying == VARYING_SLOT_PSIZ) {
- assert(linked_xfb_info->Outputs[i].NumComponents == 1);
+ assert(components == 1);
component_mask <<= 3;
+ } else if (varying == VARYING_SLOT_LAYER) {
+ assert(components == 1);
+ component_mask <<= 1;
+ } else if (varying == VARYING_SLOT_VIEWPORT) {
+ assert(components == 1);
+ component_mask <<= 2;
} else {
component_mask <<= linked_xfb_info->Outputs[i].ComponentOffset;
}
- buffer_mask |= 1 << buffer;
+ buffer_mask[stream_id] |= 1 << buffer;
decl |= buffer << SO_DECL_OUTPUT_BUFFER_SLOT_SHIFT;
- decl |= vue_map->varying_to_slot[varying] <<
- SO_DECL_REGISTER_INDEX_SHIFT;
+ if (varying == VARYING_SLOT_LAYER || varying == VARYING_SLOT_VIEWPORT) {
+ decl |= vue_map->varying_to_slot[VARYING_SLOT_PSIZ] <<
+ SO_DECL_REGISTER_INDEX_SHIFT;
+ } else {
+ assert(vue_map->varying_to_slot[varying] >= 0);
+ decl |= vue_map->varying_to_slot[varying] <<
+ SO_DECL_REGISTER_INDEX_SHIFT;
+ }
decl |= component_mask << SO_DECL_COMPONENT_MASK_SHIFT;
- /* This assert should be true until GL_ARB_transform_feedback_instanced
- * is added and we start using the hole flag.
+ /* Mesa doesn't store entries for gl_SkipComponents in the Outputs[]
+ * array. Instead, it simply increments DstOffset for the following
+ * input by the number of components that should be skipped.
+ *
+ * Our hardware is unusual in that it requires us to program SO_DECLs
+ * for fake "hole" components, rather than simply taking the offset
+ * for each real varying. Each hole can have size 1, 2, 3, or 4; we
+ * program as many size = 4 holes as we can, then a final hole to
+ * accommodate the final 1, 2, or 3 remaining.
*/
+ int skip_components =
+ linked_xfb_info->Outputs[i].DstOffset - next_offset[buffer];
+
+ next_offset[buffer] += skip_components;
+
+ while (skip_components >= 4) {
+ so_decl[stream_id][decls[stream_id]++] = SO_DECL_HOLE_FLAG | 0xf;
+ skip_components -= 4;
+ }
+ if (skip_components > 0)
+ so_decl[stream_id][decls[stream_id]++] =
+ SO_DECL_HOLE_FLAG | ((1 << skip_components) - 1);
+
assert(linked_xfb_info->Outputs[i].DstOffset == next_offset[buffer]);
- next_offset[buffer] += linked_xfb_info->Outputs[i].NumComponents;
+ next_offset[buffer] += components;
- so_decl[i] = decl;
+ so_decl[stream_id][decls[stream_id]++] = decl;
+
+ if (decls[stream_id] > max_decls)
+ max_decls = decls[stream_id];
}
- BEGIN_BATCH(linked_xfb_info->NumOutputs * 2 + 3);
- OUT_BATCH(_3DSTATE_SO_DECL_LIST << 16 |
- (linked_xfb_info->NumOutputs * 2 + 1));
+ BEGIN_BATCH(max_decls * 2 + 3);
+ OUT_BATCH(_3DSTATE_SO_DECL_LIST << 16 | (max_decls * 2 + 1));
- OUT_BATCH((buffer_mask << SO_STREAM_TO_BUFFER_SELECTS_0_SHIFT) |
- (0 << SO_STREAM_TO_BUFFER_SELECTS_1_SHIFT) |
- (0 << SO_STREAM_TO_BUFFER_SELECTS_2_SHIFT) |
- (0 << SO_STREAM_TO_BUFFER_SELECTS_3_SHIFT));
+ OUT_BATCH((buffer_mask[0] << SO_STREAM_TO_BUFFER_SELECTS_0_SHIFT) |
+ (buffer_mask[1] << SO_STREAM_TO_BUFFER_SELECTS_1_SHIFT) |
+ (buffer_mask[2] << SO_STREAM_TO_BUFFER_SELECTS_2_SHIFT) |
+ (buffer_mask[3] << SO_STREAM_TO_BUFFER_SELECTS_3_SHIFT));
- OUT_BATCH((linked_xfb_info->NumOutputs << SO_NUM_ENTRIES_0_SHIFT) |
- (0 << SO_NUM_ENTRIES_1_SHIFT) |
- (0 << SO_NUM_ENTRIES_2_SHIFT) |
- (0 << SO_NUM_ENTRIES_3_SHIFT));
+ OUT_BATCH((decls[0] << SO_NUM_ENTRIES_0_SHIFT) |
+ (decls[1] << SO_NUM_ENTRIES_1_SHIFT) |
+ (decls[2] << SO_NUM_ENTRIES_2_SHIFT) |
+ (decls[3] << SO_NUM_ENTRIES_3_SHIFT));
- for (i = 0; i < linked_xfb_info->NumOutputs; i++) {
- OUT_BATCH(so_decl[i]);
- OUT_BATCH(0);
+ for (int i = 0; i < max_decls; i++) {
+ /* Stream 1 | Stream 0 */
+ OUT_BATCH(((uint32_t) so_decl[1][i]) << 16 | so_decl[0][i]);
+ /* Stream 3 | Stream 2 */
+ OUT_BATCH(((uint32_t) so_decl[3][i]) << 16 | so_decl[2][i]);
}
ADVANCE_BATCH();
upload_3dstate_streamout(struct brw_context *brw, bool active,
const struct brw_vue_map *vue_map)
{
- struct intel_context *intel = &brw->intel;
- struct gl_context *ctx = &intel->ctx;
+ struct gl_context *ctx = &brw->ctx;
/* BRW_NEW_TRANSFORM_FEEDBACK */
struct gl_transform_feedback_object *xfb_obj =
ctx->TransformFeedback.CurrentObject;
* point by reading less and offsetting the register index in the
* SO_DECLs.
*/
- dw2 |= urb_entry_read_offset << SO_STREAM_0_VERTEX_READ_OFFSET_SHIFT;
- dw2 |= (urb_entry_read_length - 1) <<
- SO_STREAM_0_VERTEX_READ_LENGTH_SHIFT;
+ dw2 |= SET_FIELD(urb_entry_read_offset, SO_STREAM_0_VERTEX_READ_OFFSET);
+ dw2 |= SET_FIELD(urb_entry_read_length - 1, SO_STREAM_0_VERTEX_READ_LENGTH);
+
+ dw2 |= SET_FIELD(urb_entry_read_offset, SO_STREAM_1_VERTEX_READ_OFFSET);
+ dw2 |= SET_FIELD(urb_entry_read_length - 1, SO_STREAM_1_VERTEX_READ_LENGTH);
+
+ dw2 |= SET_FIELD(urb_entry_read_offset, SO_STREAM_2_VERTEX_READ_OFFSET);
+ dw2 |= SET_FIELD(urb_entry_read_length - 1, SO_STREAM_2_VERTEX_READ_LENGTH);
+
+ dw2 |= SET_FIELD(urb_entry_read_offset, SO_STREAM_3_VERTEX_READ_OFFSET);
+ dw2 |= SET_FIELD(urb_entry_read_length - 1, SO_STREAM_3_VERTEX_READ_LENGTH);
}
BEGIN_BATCH(3);
static void
upload_sol_state(struct brw_context *brw)
{
- struct intel_context *intel = &brw->intel;
- struct gl_context *ctx = &intel->ctx;
+ struct gl_context *ctx = &brw->ctx;
/* BRW_NEW_TRANSFORM_FEEDBACK */
bool active = _mesa_is_xfb_active_and_unpaused(ctx);
if (active) {
upload_3dstate_so_buffers(brw);
/* BRW_NEW_VUE_MAP_GEOM_OUT */
- upload_3dstate_so_decl_list(brw, &brw->vue_map_geom_out);
-
- /* If we don't have hardware contexts, then some other client may have
- * changed the SO write offsets, and we need to rewrite them.
- */
- if (!intel->hw_ctx)
- intel->batch.needs_sol_reset = true;
+ gen7_upload_3dstate_so_decl_list(brw, &brw->vue_map_geom_out);
}
/* Finally, set up the SOL stage. This command must always follow updates to
const struct brw_tracked_state gen7_sol_state = {
.dirty = {
- .mesa = (_NEW_LIGHT),
- .brw = (BRW_NEW_BATCH |
- BRW_NEW_VERTEX_PROGRAM |
- BRW_NEW_VUE_MAP_GEOM_OUT |
- BRW_NEW_TRANSFORM_FEEDBACK)
+ .mesa = _NEW_LIGHT,
+ .brw = BRW_NEW_BATCH |
+ BRW_NEW_VUE_MAP_GEOM_OUT |
+ BRW_NEW_TRANSFORM_FEEDBACK,
},
.emit = upload_sol_state,
};
+/**
+ * Tally the number of primitives generated so far.
+ *
+ * The buffer contains a series of pairs:
+ * (<start0, start1, start2, start3>, <end0, end1, end2, end3>) ;
+ * (<start0, start1, start2, start3>, <end0, end1, end2, end3>) ;
+ *
+ * For each stream, we subtract the pair of values (end - start) to get the
+ * number of primitives generated during one section. We accumulate these
+ * values, adding them up to get the total number of primitives generated.
+ */
+static void
+gen7_tally_prims_generated(struct brw_context *brw,
+ struct brw_transform_feedback_object *obj)
+{
+ /* If the current batch is still contributing to the number of primitives
+ * generated, flush it now so the results will be present when mapped.
+ */
+ if (drm_intel_bo_references(brw->batch.bo, obj->prim_count_bo))
+ intel_batchbuffer_flush(brw);
+
+ if (unlikely(brw->perf_debug && drm_intel_bo_busy(obj->prim_count_bo)))
+ perf_debug("Stalling for # of transform feedback primitives written.\n");
+
+ drm_intel_bo_map(obj->prim_count_bo, false);
+ uint64_t *prim_counts = obj->prim_count_bo->virtual;
+
+ assert(obj->prim_count_buffer_index % (2 * BRW_MAX_XFB_STREAMS) == 0);
+ int pairs = obj->prim_count_buffer_index / (2 * BRW_MAX_XFB_STREAMS);
+
+ for (int i = 0; i < pairs; i++) {
+ for (int s = 0; s < BRW_MAX_XFB_STREAMS; s++) {
+ obj->prims_generated[s] +=
+ prim_counts[BRW_MAX_XFB_STREAMS + s] - prim_counts[s];
+ }
+ prim_counts += 2 * BRW_MAX_XFB_STREAMS; /* move to the next pair */
+ }
+
+ drm_intel_bo_unmap(obj->prim_count_bo);
+
+ /* We've already gathered up the old data; we can safely overwrite it now. */
+ obj->prim_count_buffer_index = 0;
+}
+
+/**
+ * Store the SO_NUM_PRIMS_WRITTEN counters for each stream (4 uint64_t values)
+ * to prim_count_bo.
+ *
+ * If prim_count_bo is out of space, gather up the results so far into
+ * prims_generated[] and allocate a new buffer with enough space.
+ *
+ * The number of primitives written is used to compute the number of vertices
+ * written to a transform feedback stream, which is required to implement
+ * DrawTransformFeedback().
+ */
+static void
+gen7_save_primitives_written_counters(struct brw_context *brw,
+ struct brw_transform_feedback_object *obj)
+{
+ const int streams = BRW_MAX_XFB_STREAMS;
+
+ /* Check if there's enough space for a new pair of four values. */
+ if (obj->prim_count_bo != NULL &&
+ obj->prim_count_buffer_index + 2 * streams >= 4096 / sizeof(uint64_t)) {
+ /* Gather up the results so far and release the BO. */
+ gen7_tally_prims_generated(brw, obj);
+ }
+
+ /* Flush any drawing so that the counters have the right values. */
+ brw_emit_mi_flush(brw);
+
+ /* Emit MI_STORE_REGISTER_MEM commands to write the values. */
+ for (int i = 0; i < streams; i++) {
+ brw_store_register_mem64(brw, obj->prim_count_bo,
+ GEN7_SO_NUM_PRIMS_WRITTEN(i),
+ obj->prim_count_buffer_index + i);
+ }
+
+ /* Update where to write data to. */
+ obj->prim_count_buffer_index += streams;
+}
+
+/**
+ * Compute the number of vertices written by this transform feedback operation.
+ */
+static void
+brw_compute_xfb_vertices_written(struct brw_context *brw,
+ struct brw_transform_feedback_object *obj)
+{
+ if (obj->vertices_written_valid || !obj->base.EndedAnytime)
+ return;
+
+ unsigned vertices_per_prim = 0;
+
+ switch (obj->primitive_mode) {
+ case GL_POINTS:
+ vertices_per_prim = 1;
+ break;
+ case GL_LINES:
+ vertices_per_prim = 2;
+ break;
+ case GL_TRIANGLES:
+ vertices_per_prim = 3;
+ break;
+ default:
+ unreachable("Invalid transform feedback primitive mode.");
+ }
+
+ /* Get the number of primitives generated. */
+ gen7_tally_prims_generated(brw, obj);
+
+ for (int i = 0; i < BRW_MAX_XFB_STREAMS; i++) {
+ obj->vertices_written[i] = vertices_per_prim * obj->prims_generated[i];
+ }
+ obj->vertices_written_valid = true;
+}
+
+/**
+ * GetTransformFeedbackVertexCount() driver hook.
+ *
+ * Returns the number of vertices written to a particular stream by the last
+ * Begin/EndTransformFeedback block. Used to implement DrawTransformFeedback().
+ */
+GLsizei
+brw_get_transform_feedback_vertex_count(struct gl_context *ctx,
+ struct gl_transform_feedback_object *obj,
+ GLuint stream)
+{
+ struct brw_context *brw = brw_context(ctx);
+ struct brw_transform_feedback_object *brw_obj =
+ (struct brw_transform_feedback_object *) obj;
+
+ assert(obj->EndedAnytime);
+ assert(stream < BRW_MAX_XFB_STREAMS);
+
+ brw_compute_xfb_vertices_written(brw, brw_obj);
+ return brw_obj->vertices_written[stream];
+}
+
+void
+gen7_begin_transform_feedback(struct gl_context *ctx, GLenum mode,
+ struct gl_transform_feedback_object *obj)
+{
+ struct brw_context *brw = brw_context(ctx);
+ struct brw_transform_feedback_object *brw_obj =
+ (struct brw_transform_feedback_object *) obj;
+
+ /* Reset the SO buffer offsets to 0. */
+ if (brw->gen >= 8) {
+ brw_obj->zero_offsets = true;
+ } else {
+ intel_batchbuffer_flush(brw);
+ brw->batch.needs_sol_reset = true;
+ }
+
+ /* We're about to lose the information needed to compute the number of
+ * vertices written during the last Begin/EndTransformFeedback section,
+ * so we can't delay it any further.
+ */
+ brw_compute_xfb_vertices_written(brw, brw_obj);
+
+ /* No primitives have been generated yet. */
+ for (int i = 0; i < BRW_MAX_XFB_STREAMS; i++) {
+ brw_obj->prims_generated[i] = 0;
+ }
+
+ /* Store the starting value of the SO_NUM_PRIMS_WRITTEN counters. */
+ gen7_save_primitives_written_counters(brw, brw_obj);
+
+ brw_obj->primitive_mode = mode;
+}
+
void
gen7_end_transform_feedback(struct gl_context *ctx,
struct gl_transform_feedback_object *obj)
{
- /* Because we have to rely on the kernel to reset our SO write offsets, and
- * we only get to do it once per batchbuffer, flush the batch after feedback
- * so another transform feedback can get the write offset reset it needs.
- *
- * This also covers any cache flushing required.
+ /* After EndTransformFeedback, it's likely that the client program will try
+ * to draw using the contents of the transform feedback buffer as vertex
+ * input. In order for this to work, we need to flush the data through at
+ * least the GS stage of the pipeline, and flush out the render cache. For
+ * simplicity, just do a full flush.
+ */
+ struct brw_context *brw = brw_context(ctx);
+ struct brw_transform_feedback_object *brw_obj =
+ (struct brw_transform_feedback_object *) obj;
+
+ /* Store the ending value of the SO_NUM_PRIMS_WRITTEN counters. */
+ gen7_save_primitives_written_counters(brw, brw_obj);
+
+ /* EndTransformFeedback() means that we need to update the number of
+ * vertices written. Since it's only necessary if DrawTransformFeedback()
+ * is called and it means mapping a buffer object, we delay computing it
+ * until it's absolutely necessary to try and avoid stalls.
+ */
+ brw_obj->vertices_written_valid = false;
+}
+
+void
+gen7_pause_transform_feedback(struct gl_context *ctx,
+ struct gl_transform_feedback_object *obj)
+{
+ struct brw_context *brw = brw_context(ctx);
+ struct brw_transform_feedback_object *brw_obj =
+ (struct brw_transform_feedback_object *) obj;
+
+ /* Flush any drawing so that the counters have the right values. */
+ brw_emit_mi_flush(brw);
+
+ /* Save the SOL buffer offset register values. */
+ if (brw->gen < 8) {
+ for (int i = 0; i < 4; i++) {
+ BEGIN_BATCH(3);
+ OUT_BATCH(MI_STORE_REGISTER_MEM | (3 - 2));
+ OUT_BATCH(GEN7_SO_WRITE_OFFSET(i));
+ OUT_RELOC(brw_obj->offset_bo,
+ I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+ i * sizeof(uint32_t));
+ ADVANCE_BATCH();
+ }
+ }
+
+ /* Store the temporary ending value of the SO_NUM_PRIMS_WRITTEN counters.
+ * While this operation is paused, other transform feedback actions may
+ * occur, which will contribute to the counters. We need to exclude that
+ * from our counts.
*/
+ gen7_save_primitives_written_counters(brw, brw_obj);
+}
+
+void
+gen7_resume_transform_feedback(struct gl_context *ctx,
+ struct gl_transform_feedback_object *obj)
+{
struct brw_context *brw = brw_context(ctx);
- struct intel_context *intel = &brw->intel;
+ struct brw_transform_feedback_object *brw_obj =
+ (struct brw_transform_feedback_object *) obj;
+
+ /* Reload the SOL buffer offset registers. */
+ if (brw->gen < 8) {
+ for (int i = 0; i < 4; i++) {
+ BEGIN_BATCH(3);
+ OUT_BATCH(GEN7_MI_LOAD_REGISTER_MEM | (3 - 2));
+ OUT_BATCH(GEN7_SO_WRITE_OFFSET(i));
+ OUT_RELOC(brw_obj->offset_bo,
+ I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
+ i * sizeof(uint32_t));
+ ADVANCE_BATCH();
+ }
+ }
- intel_batchbuffer_flush(intel);
+ /* Store the new starting value of the SO_NUM_PRIMS_WRITTEN counters. */
+ gen7_save_primitives_written_counters(brw, brw_obj);
}