From: Marek Olšák <marek.ol...@amd.com>
---
src/gallium/drivers/radeonsi/si_shader.h | 1 +
src/gallium/drivers/radeonsi/si_state_shaders.c | 40 +++++++++++++++----------
2 files changed, 25 insertions(+), 16 deletions(-)
diff --git a/src/gallium/drivers/radeonsi/si_shader.h
b/src/gallium/drivers/radeonsi/si_shader.h
index aab902b..7112621 100644
--- a/src/gallium/drivers/radeonsi/si_shader.h
+++ b/src/gallium/drivers/radeonsi/si_shader.h
@@ -329,20 +329,21 @@ struct si_shader_selector {
/* GS parameters. */
unsigned esgs_itemsize;
unsigned gs_input_verts_per_prim;
unsigned gs_output_prim;
unsigned gs_max_out_vertices;
unsigned gs_num_invocations;
unsigned max_gs_stream; /* count - 1 */
unsigned gsvs_vertex_size;
unsigned max_gsvs_emit_size;
+ unsigned enabled_streamout_buffer_mask;
/* PS parameters. */
unsigned color_attr_index[2];
unsigned db_shader_control;
/* Set 0xf or 0x0 (4 bits) per each written output.
* ANDed with spi_shader_col_format.
*/
unsigned colors_written_4bit;
/* CS parameters */
diff --git a/src/gallium/drivers/radeonsi/si_state_shaders.c
b/src/gallium/drivers/radeonsi/si_state_shaders.c
index 55e881c..5cbb91b 100644
--- a/src/gallium/drivers/radeonsi/si_state_shaders.c
+++ b/src/gallium/drivers/radeonsi/si_state_shaders.c
@@ -1949,20 +1949,27 @@ static void *si_create_shader_selector(struct
pipe_context *ctx,
}
sel->so = state->stream_output;
tgsi_scan_shader(state->tokens, &sel->info);
sel->type = sel->info.processor;
p_atomic_inc(&sscreen->b.num_shaders_created);
si_get_active_slot_masks(&sel->info,
&sel->active_const_and_shader_buffers,
&sel->active_samplers_and_images);
+ /* Record which streamout buffers are enabled. */
+ for (i = 0; i < sel->so.num_outputs; i++) {
+ sel->enabled_streamout_buffer_mask |=
+ (1 << sel->so.output[i].output_buffer) <<
+ (sel->so.output[i].stream * 4);
+ }
+
/* The prolog is a no-op if there are no inputs. */
sel->vs_needs_prolog = sel->type == PIPE_SHADER_VERTEX &&
sel->info.num_inputs;
/* Set which opcode uses which (i,j) pair. */
if (sel->info.uses_persp_opcode_interp_centroid)
sel->info.uses_persp_centroid = true;
if (sel->info.uses_linear_opcode_interp_centroid)
sel->info.uses_linear_centroid = true;
@@ -2130,34 +2137,49 @@ static void *si_create_shader_selector(struct
pipe_context *ctx,
r600_can_dump_shader(&sscreen->b, sel->info.processor))
si_init_shader_selector_async(sel, -1);
else
util_queue_add_job(&sscreen->shader_compiler_queue, sel,
&sel->ready, si_init_shader_selector_async,
NULL);
return sel;
}
+static void si_update_streamout_state(struct si_context *sctx)
+{
+ struct si_shader_selector *shader_with_so =
+ sctx->gs_shader.cso ? sctx->gs_shader.cso :
+ sctx->tes_shader.cso ? sctx->tes_shader.cso :
+ sctx->vs_shader.cso;
+ if (!shader_with_so)
+ return;
+
+ sctx->b.streamout.enabled_stream_buffers_mask =
+ shader_with_so->enabled_streamout_buffer_mask;
+ sctx->b.streamout.stride_in_dw = shader_with_so->so.stride;
+}
+
static void si_bind_vs_shader(struct pipe_context *ctx, void *state)
{
struct si_context *sctx = (struct si_context *)ctx;
struct si_shader_selector *sel = state;
if (sctx->vs_shader.cso == sel)
return;
sctx->vs_shader.cso = sel;
sctx->vs_shader.current = sel ? sel->first_variant : NULL;
sctx->do_update_shaders = true;
si_mark_atom_dirty(sctx, &sctx->clip_regs);
r600_update_vs_writes_viewport_index(&sctx->b, si_get_vs_info(sctx));
si_set_active_descriptors_for_shader(sctx, sel);
+ si_update_streamout_state(sctx);
}
static void si_update_tess_uses_prim_id(struct si_context *sctx)
{
sctx->ia_multi_vgt_param_key.u.tess_uses_prim_id =
(sctx->tes_shader.cso &&
sctx->tes_shader.cso->info.uses_primid) ||
(sctx->tcs_shader.cso &&
sctx->tcs_shader.cso->info.uses_primid) ||
(sctx->gs_shader.cso &&
@@ -2182,20 +2204,21 @@ static void si_bind_gs_shader(struct pipe_context *ctx,
void *state)
si_mark_atom_dirty(sctx, &sctx->clip_regs);
sctx->last_rast_prim = -1; /* reset this so that it gets updated */
if (enable_changed) {
si_shader_change_notify(sctx);
if (sctx->ia_multi_vgt_param_key.u.uses_tess)
si_update_tess_uses_prim_id(sctx);
}
r600_update_vs_writes_viewport_index(&sctx->b, si_get_vs_info(sctx));
si_set_active_descriptors_for_shader(sctx, sel);
+ si_update_streamout_state(sctx);
}
static void si_bind_tcs_shader(struct pipe_context *ctx, void *state)
{
struct si_context *sctx = (struct si_context *)ctx;
struct si_shader_selector *sel = state;
bool enable_changed = !!sctx->tcs_shader.cso != !!sel;
if (sctx->tcs_shader.cso == sel)
return;
@@ -2227,20 +2250,21 @@ static void si_bind_tes_shader(struct pipe_context
*ctx, void *state)
sctx->do_update_shaders = true;
si_mark_atom_dirty(sctx, &sctx->clip_regs);
sctx->last_rast_prim = -1; /* reset this so that it gets updated */
if (enable_changed) {
si_shader_change_notify(sctx);
sctx->last_tes_sh_base = -1; /* invalidate derived tess state */
}
r600_update_vs_writes_viewport_index(&sctx->b, si_get_vs_info(sctx));
si_set_active_descriptors_for_shader(sctx, sel);
+ si_update_streamout_state(sctx);
}
static void si_bind_ps_shader(struct pipe_context *ctx, void *state)
{
struct si_context *sctx = (struct si_context *)ctx;
struct si_shader_selector *sel = state;
/* skip if supplied shader is one already in use */
if (sctx->ps_shader.cso == sel)
return;
@@ -2984,32 +3008,20 @@ static void si_update_vgt_shader_config(struct
si_context *sctx)
}
if (sctx->b.chip_class >= GFX9)
stages |= S_028B54_MAX_PRIMGRP_IN_WAVE(2);
si_pm4_set_reg(*pm4, R_028B54_VGT_SHADER_STAGES_EN, stages);
}
si_pm4_bind_state(sctx, vgt_shader_config, *pm4);
}
-static void si_update_so(struct si_context *sctx, struct si_shader_selector
*shader)
-{
- struct pipe_stream_output_info *so = &shader->so;
- uint32_t enabled_stream_buffers_mask = 0;
- int i;
-
- for (i = 0; i < so->num_outputs; i++)
- enabled_stream_buffers_mask |= (1 <<
so->output[i].output_buffer) << (so->output[i].stream * 4);
- sctx->b.streamout.enabled_stream_buffers_mask =
enabled_stream_buffers_mask;
- sctx->b.streamout.stride_in_dw = shader->so.stride;
-}
-
bool si_update_shaders(struct si_context *sctx)
{
struct pipe_context *ctx = (struct pipe_context*)sctx;
struct si_compiler_ctx_state compiler_state;
struct si_state_rasterizer *rs = sctx->queued.named.rasterizer;
struct si_shader *old_vs = si_get_vs_state(sctx);
bool old_clip_disable = old_vs ? old_vs->key.opt.hw_vs.clip_disable :
false;
int r;
compiler_state.tm = sctx->tm;
@@ -3063,54 +3075,50 @@ bool si_update_shaders(struct si_context *sctx)
return false;
si_pm4_bind_state(sctx, es,
sctx->tes_shader.current->pm4);
}
} else {
/* TES as VS */
r = si_shader_select(ctx, &sctx->tes_shader,
&compiler_state);
if (r)
return false;
si_pm4_bind_state(sctx, vs,
sctx->tes_shader.current->pm4);
- si_update_so(sctx, sctx->tes_shader.cso);
}
} else if (sctx->gs_shader.cso) {
if (sctx->b.chip_class <= VI) {
/* VS as ES */
r = si_shader_select(ctx, &sctx->vs_shader,
&compiler_state);
if (r)
return false;
si_pm4_bind_state(sctx, es,
sctx->vs_shader.current->pm4);
si_pm4_bind_state(sctx, ls, NULL);
si_pm4_bind_state(sctx, hs, NULL);
}
} else {
/* VS as VS */
r = si_shader_select(ctx, &sctx->vs_shader, &compiler_state);
if (r)
return false;
si_pm4_bind_state(sctx, vs, sctx->vs_shader.current->pm4);
- si_update_so(sctx, sctx->vs_shader.cso);
-
si_pm4_bind_state(sctx, ls, NULL);
si_pm4_bind_state(sctx, hs, NULL);
}
/* Update GS. */
if (sctx->gs_shader.cso) {
r = si_shader_select(ctx, &sctx->gs_shader, &compiler_state);
if (r)
return false;
si_pm4_bind_state(sctx, gs, sctx->gs_shader.current->pm4);
si_pm4_bind_state(sctx, vs,
sctx->gs_shader.cso->gs_copy_shader->pm4);
- si_update_so(sctx, sctx->gs_shader.cso);
if (!si_update_gs_ring_buffers(sctx))
return false;
} else {
si_pm4_bind_state(sctx, gs, NULL);
if (sctx->b.chip_class <= VI)
si_pm4_bind_state(sctx, es, NULL);
}
si_update_vgt_shader_config(sctx);
--
2.7.4
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
