summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorKenneth Graunke <[email protected]>2018-07-11 12:45:19 -0700
committerKenneth Graunke <[email protected]>2019-02-21 10:26:07 -0800
commitcef0b8b13bbcc4b95c4ed21b918e0126d273f79d (patch)
treec1d29c6af5de8245079ed93289e5575828a860bb
parent059c096eff1a380fe72be675893a401d6c7dfa1a (diff)
iris: streamout
-rw-r--r--src/gallium/drivers/iris/iris_context.h11
-rw-r--r--src/gallium/drivers/iris/iris_program.c6
-rw-r--r--src/gallium/drivers/iris/iris_program_cache.c10
-rw-r--r--src/gallium/drivers/iris/iris_state.c84
4 files changed, 88 insertions, 23 deletions
diff --git a/src/gallium/drivers/iris/iris_context.h b/src/gallium/drivers/iris/iris_context.h
index ca8596623e1..24d891f97fd 100644
--- a/src/gallium/drivers/iris/iris_context.h
+++ b/src/gallium/drivers/iris/iris_context.h
@@ -171,9 +171,10 @@ struct iris_compiled_shader {
struct brw_stage_prog_data *prog_data;
/**
- * Derived 3DSTATE_SO_DECL_LIST packet (for transform feedback).
+ * Derived 3DSTATE_STREAMOUT and 3DSTATE_SO_DECL_LIST packets
+ * (the VUE-based information for transform feedback outputs).
*/
- uint32_t *so_decl_list;
+ uint32_t *streamout;
/**
* Shader packets and other data derived from prog_data. These must be
@@ -284,7 +285,9 @@ struct iris_context {
unsigned num_samplers[MESA_SHADER_STAGES];
unsigned num_textures[MESA_SHADER_STAGES];
- uint32_t *so_decl_list;
+ /** 3DSTATE_STREAMOUT and 3DSTATE_SO_DECL_LIST packets */
+ bool streamout_active;
+ uint32_t *streamout;
struct iris_state_ref unbound_tex;
@@ -392,7 +395,7 @@ void iris_upload_and_bind_shader(struct iris_context *ice,
const void *key,
const void *assembly,
struct brw_stage_prog_data *prog_data,
- uint32_t *so_decl_list);
+ uint32_t *streamout);
const void *iris_find_previous_compile(const struct iris_context *ice,
enum iris_program_cache_id cache_id,
unsigned program_string_id);
diff --git a/src/gallium/drivers/iris/iris_program.c b/src/gallium/drivers/iris/iris_program.c
index 35fbda92c9e..b1bfce0dd25 100644
--- a/src/gallium/drivers/iris/iris_program.c
+++ b/src/gallium/drivers/iris/iris_program.c
@@ -577,9 +577,9 @@ iris_update_compiled_shaders(struct iris_context *ice)
struct iris_compiled_shader *shader = last_vue_shader(ice);
update_last_vue_map(ice, shader->prog_data);
- if (ice->state.so_decl_list != shader->so_decl_list) {
- ice->state.so_decl_list = shader->so_decl_list;
- ice->state.dirty |= IRIS_DIRTY_SO_DECL_LIST;
+ if (ice->state.streamout != shader->streamout) {
+ ice->state.streamout = shader->streamout;
+ ice->state.dirty |= IRIS_DIRTY_SO_DECL_LIST | IRIS_DIRTY_STREAMOUT;
}
if (dirty & IRIS_DIRTY_UNCOMPILED_FS)
diff --git a/src/gallium/drivers/iris/iris_program_cache.c b/src/gallium/drivers/iris/iris_program_cache.c
index c2323bd0995..0bf01205ffb 100644
--- a/src/gallium/drivers/iris/iris_program_cache.c
+++ b/src/gallium/drivers/iris/iris_program_cache.c
@@ -212,7 +212,7 @@ iris_upload_shader(struct iris_context *ice,
const void *key,
const void *assembly,
struct brw_stage_prog_data *prog_data,
- uint32_t *so_decl_list)
+ uint32_t *streamout)
{
struct iris_screen *screen = (void *) ice->ctx.screen;
struct gen_device_info *devinfo = &screen->devinfo;
@@ -242,12 +242,12 @@ iris_upload_shader(struct iris_context *ice,
}
shader->prog_data = prog_data;
- shader->so_decl_list = so_decl_list;
+ shader->streamout = streamout;
ralloc_steal(shader, shader->prog_data);
ralloc_steal(shader->prog_data, prog_data->param);
ralloc_steal(shader->prog_data, prog_data->pull_param);
- ralloc_steal(shader, shader->so_decl_list);
+ ralloc_steal(shader, shader->streamout);
/* Store the 3DSTATE shader packets and other derived state. */
ice->vtbl.store_derived_program_state(devinfo, cache_id, shader);
@@ -269,13 +269,13 @@ iris_upload_and_bind_shader(struct iris_context *ice,
const void *key,
const void *assembly,
struct brw_stage_prog_data *prog_data,
- uint32_t *so_decl_list)
+ uint32_t *streamout)
{
assert(cache_id != IRIS_CACHE_BLORP);
struct iris_compiled_shader *shader =
iris_upload_shader(ice, cache_id, key_size_for_cache(cache_id), key,
- assembly, prog_data, so_decl_list);
+ assembly, prog_data, streamout);
ice->shaders.prog[cache_id] = shader;
ice->state.dirty |= dirty_flag_for_cache(cache_id);
diff --git a/src/gallium/drivers/iris/iris_state.c b/src/gallium/drivers/iris/iris_state.c
index 9d1f3d9dcf5..b6c90ad49fa 100644
--- a/src/gallium/drivers/iris/iris_state.c
+++ b/src/gallium/drivers/iris/iris_state.c
@@ -433,6 +433,7 @@ struct iris_genx_state {
struct iris_depth_buffer_state depth_buffer;
uint32_t so_buffers[4 * GENX(3DSTATE_SO_BUFFER_length)];
+ uint32_t streamout[4 * GENX(3DSTATE_STREAMOUT_length)];
};
static void
@@ -626,6 +627,7 @@ struct iris_rasterizer_state {
uint32_t line_stipple[GENX(3DSTATE_LINE_STIPPLE_length)];
bool flatshade; /* for shader state */
+ bool flatshade_first; /* for stream output */
bool clamp_fragment_color; /* for shader state */
bool light_twoside; /* for shader state */
bool rasterizer_discard; /* for 3DSTATE_STREAMOUT */
@@ -657,6 +659,7 @@ iris_create_rasterizer_state(struct pipe_context *ctx,
#endif
cso->flatshade = state->flatshade;
+ cso->flatshade_first = state->flatshade_first;
cso->clamp_fragment_color = state->clamp_fragment_color;
cso->light_twoside = state->light_twoside;
cso->rasterizer_discard = state->rasterizer_discard;
@@ -774,7 +777,7 @@ iris_bind_rasterizer_state(struct pipe_context *ctx, void *state)
if (cso_changed(line_stipple_enable) || cso_changed(poly_stipple_enable))
ice->state.dirty |= IRIS_DIRTY_WM;
- if (cso_changed(rasterizer_discard))
+ if (cso_changed(rasterizer_discard) || cso_changed(flatshade_first))
ice->state.dirty |= IRIS_DIRTY_STREAMOUT;
}
@@ -1696,7 +1699,18 @@ iris_set_stream_output_targets(struct pipe_context *ctx,
const unsigned *offsets)
{
struct iris_context *ice = (struct iris_context *) ctx;
- uint32_t *so_buffers = ice->state.genx->so_buffers;
+ struct iris_genx_state *genx = ice->state.genx;
+ uint32_t *so_buffers = genx->so_buffers;
+
+ const bool active = num_targets > 0;
+ if (ice->state.streamout_active != active) {
+ ice->state.streamout_active = active;
+ ice->state.dirty |= IRIS_DIRTY_STREAMOUT;
+ }
+
+ /* No need to update 3DSTATE_SO_BUFFER unless SOL is active. */
+ if (!active)
+ return;
for (unsigned i = 0; i < 4; i++,
so_buffers += GENX(3DSTATE_SO_BUFFER_length)) {
@@ -1792,9 +1806,36 @@ iris_create_so_decl_list(const struct pipe_stream_output_info *info,
max_decls = decls[stream_id];
}
- uint32_t *dw = ralloc_size(NULL, sizeof(uint32_t) * (3 + 2 * max_decls));
+ unsigned dwords = GENX(3DSTATE_STREAMOUT_length) + (3 + 2 * max_decls);
+ uint32_t *map = ralloc_size(NULL, sizeof(uint32_t) * dwords);
+ uint32_t *so_decl_map = map + GENX(3DSTATE_STREAMOUT_length);
+
+ iris_pack_command(GENX(3DSTATE_STREAMOUT), map, sol) {
+ int urb_entry_read_offset = 0;
+ int urb_entry_read_length = (vue_map->num_slots + 1) / 2 -
+ urb_entry_read_offset;
- iris_pack_command(GENX(3DSTATE_SO_DECL_LIST), dw, list) {
+ /* We always read the whole vertex. This could be reduced at some
+ * point by reading less and offsetting the register index in the
+ * SO_DECLs.
+ */
+ sol.Stream0VertexReadOffset = urb_entry_read_offset;
+ sol.Stream0VertexReadLength = urb_entry_read_length - 1;
+ sol.Stream1VertexReadOffset = urb_entry_read_offset;
+ sol.Stream1VertexReadLength = urb_entry_read_length - 1;
+ sol.Stream2VertexReadOffset = urb_entry_read_offset;
+ sol.Stream2VertexReadLength = urb_entry_read_length - 1;
+ sol.Stream3VertexReadOffset = urb_entry_read_offset;
+ sol.Stream3VertexReadLength = urb_entry_read_length - 1;
+
+ /* Set buffer pitches; 0 means unbound. */
+ sol.Buffer0SurfacePitch = 4 * info->stride[0];
+ sol.Buffer1SurfacePitch = 4 * info->stride[1];
+ sol.Buffer2SurfacePitch = 4 * info->stride[2];
+ sol.Buffer3SurfacePitch = 4 * info->stride[3];
+ }
+
+ iris_pack_command(GENX(3DSTATE_SO_DECL_LIST), so_decl_map, list) {
list.DWordLength = 3 + 2 * max_decls - 2;
list.StreamtoBufferSelects0 = buffer_mask[0];
list.StreamtoBufferSelects1 = buffer_mask[1];
@@ -1807,7 +1848,7 @@ iris_create_so_decl_list(const struct pipe_stream_output_info *info,
}
for (int i = 0; i < max_decls; i++) {
- iris_pack_state(GENX(SO_DECL_ENTRY), dw + 2 + i * 2, entry) {
+ iris_pack_state(GENX(SO_DECL_ENTRY), so_decl_map + 2 + i * 2, entry) {
entry.Stream0Decl = so_decl[0][i];
entry.Stream1Decl = so_decl[1][i];
entry.Stream2Decl = so_decl[2][i];
@@ -1815,7 +1856,7 @@ iris_create_so_decl_list(const struct pipe_stream_output_info *info,
}
}
- return dw;
+ return map;
}
static void
@@ -2862,13 +2903,34 @@ iris_upload_render_state(struct iris_context *ice,
4 * 4 * GENX(3DSTATE_SO_BUFFER_length));
}
- if ((dirty & IRIS_DIRTY_SO_DECL_LIST) && ice->state.so_decl_list) {
- iris_batch_emit(batch, ice->state.so_decl_list,
- 4 * ((ice->state.so_decl_list[0] & 0xff) + 2));
+ if ((dirty & IRIS_DIRTY_SO_DECL_LIST) && ice->state.streamout) {
+ uint32_t *decl_list =
+ ice->state.streamout + GENX(3DSTATE_STREAMOUT_length);
+ iris_batch_emit(batch, decl_list, 4 * ((decl_list[0] & 0xff) + 2));
}
- // XXX: SOL:
- // 3DSTATE_STREAMOUT
+ if (dirty & IRIS_DIRTY_STREAMOUT) {
+ const struct iris_rasterizer_state *cso_rast = ice->state.cso_rast;
+
+ if (!ice->state.streamout_active) {
+ iris_emit_cmd(batch, GENX(3DSTATE_STREAMOUT), sol);
+ } else {
+ uint32_t dynamic_sol[GENX(3DSTATE_STREAMOUT_length)];
+ iris_pack_command(GENX(3DSTATE_STREAMOUT), dynamic_sol, sol) {
+ sol.SOFunctionEnable = true;
+ sol.SOStatisticsEnable = true;
+
+ // XXX: GL_PRIMITIVES_GENERATED query
+ sol.RenderingDisable = cso_rast->rasterizer_discard;
+ sol.ReorderMode = cso_rast->flatshade_first ? LEADING : TRAILING;
+ }
+
+ assert(ice->state.streamout);
+
+ iris_emit_merge(batch, ice->state.streamout, dynamic_sol,
+ GENX(3DSTATE_STREAMOUT_length));
+ }
+ }
if (dirty & IRIS_DIRTY_CLIP) {
struct iris_rasterizer_state *cso_rast = ice->state.cso_rast;