switch (info->mode) {
case PIPE_PRIM_PATCHES:
return info->count / info->vertices_per_patch;
+ case PIPE_PRIM_POLYGON:
+ return info->count >= 3;
case SI_PRIM_RECTANGLE_LIST:
return info->count / 3;
default:
- return u_prims_for_vertices(info->mode, info->count);
+ return u_decomposed_prims_for_vertices(info->mode, info->count);
}
}
key->u.uses_gs)
partial_vs_wave = true;
- /* Needed for 028B6C_DISTRIBUTION_MODE != 0 */
+ /* Needed for 028B6C_DISTRIBUTION_MODE != 0. (implies >= VI) */
if (sscreen->has_distributed_tess) {
if (key->u.uses_gs) {
- if (sscreen->info.chip_class <= VI)
+ if (sscreen->info.chip_class == VI)
partial_es_wave = true;
-
- /* GPU hang workaround. */
- if (sscreen->info.family == CHIP_TONGA ||
- sscreen->info.family == CHIP_FIJI ||
- sscreen->info.family == CHIP_POLARIS10 ||
- sscreen->info.family == CHIP_POLARIS11 ||
- sscreen->info.family == CHIP_POLARIS12 ||
- sscreen->info.family == CHIP_VEGAM)
- partial_vs_wave = true;
} else {
partial_vs_wave = true;
}
* Polaris supports primitive restart with WD_SWITCH_ON_EOP=0
* for points, line strips, and tri strips.
*/
- if (sscreen->info.max_se < 4 ||
+ if (sscreen->info.max_se <= 2 ||
key->u.prim == PIPE_PRIM_POLYGON ||
key->u.prim == PIPE_PRIM_LINE_LOOP ||
key->u.prim == PIPE_PRIM_TRIANGLE_FAN ||
wd_switch_on_eop = true;
/* Required on CIK and later. */
- if (sscreen->info.max_se > 2 && !wd_switch_on_eop)
+ if (sscreen->info.max_se == 4 && !wd_switch_on_eop)
ia_switch_on_eoi = true;
+ /* HW engineers suggested that PARTIAL_VS_WAVE_ON should be set
+ * to work around a GS hang.
+ */
+ if (key->u.uses_gs &&
+ (sscreen->info.family == CHIP_TONGA ||
+ sscreen->info.family == CHIP_FIJI ||
+ sscreen->info.family == CHIP_POLARIS10 ||
+ sscreen->info.family == CHIP_POLARIS11 ||
+ sscreen->info.family == CHIP_POLARIS12 ||
+ sscreen->info.family == CHIP_VEGAM))
+ partial_vs_wave = true;
+
/* Required by Hawaii and, for some special cases, by VI. */
if (ia_switch_on_eoi &&
(sscreen->info.family == CHIP_HAWAII ||
key->u.uses_instancing)
partial_vs_wave = true;
+ /* This only applies to Polaris10 and later 4 SE chips.
+ * wd_switch_on_eop is already true on all other chips.
+ */
+ if (!wd_switch_on_eop && key->u.primitive_restart)
+ partial_vs_wave = true;
+
/* If the WD switch is false, the IA switch must be false too. */
assert(wd_switch_on_eop || !ia_switch_on_eop);
}
S_030960_EN_INST_OPT_ADV(sscreen->info.chip_class >= GFX9);
}
-void si_init_ia_multi_vgt_param_table(struct si_context *sctx)
+static void si_init_ia_multi_vgt_param_table(struct si_context *sctx)
{
for (int prim = 0; prim <= SI_PRIM_RECTANGLE_LIST; prim++)
for (int uses_instancing = 0; uses_instancing < 2; uses_instancing++)
/* Draw state. */
if (ia_multi_vgt_param != sctx->last_multi_vgt_param) {
if (sctx->chip_class >= GFX9)
- radeon_set_uconfig_reg_idx(cs, R_030960_IA_MULTI_VGT_PARAM, 4, ia_multi_vgt_param);
+ radeon_set_uconfig_reg_idx(cs, sctx->screen,
+ R_030960_IA_MULTI_VGT_PARAM, 4,
+ ia_multi_vgt_param);
else if (sctx->chip_class >= CIK)
radeon_set_context_reg_idx(cs, R_028AA8_IA_MULTI_VGT_PARAM, 1, ia_multi_vgt_param);
else
}
if (prim != sctx->last_prim) {
if (sctx->chip_class >= CIK)
- radeon_set_uconfig_reg_idx(cs, R_030908_VGT_PRIMITIVE_TYPE, 1, prim);
+ radeon_set_uconfig_reg_idx(cs, sctx->screen,
+ R_030908_VGT_PRIMITIVE_TYPE, 1, prim);
else
radeon_set_config_reg(cs, R_008958_VGT_PRIMITIVE_TYPE, prim);
}
if (sctx->chip_class >= GFX9) {
- radeon_set_uconfig_reg_idx(cs, R_03090C_VGT_INDEX_TYPE,
- 2, index_type);
+ radeon_set_uconfig_reg_idx(cs, sctx->screen,
+ R_03090C_VGT_INDEX_TYPE, 2,
+ index_type);
} else {
radeon_emit(cs, PKT3(PKT3_INDEX_TYPE, 0, 0));
radeon_emit(cs, index_type);
radeon_emit(cs, di_src_sel);
}
} else {
+ unsigned instance_count = info->instance_count;
int base_vertex;
- radeon_emit(cs, PKT3(PKT3_NUM_INSTANCES, 0, 0));
- radeon_emit(cs, info->instance_count);
+ if (sctx->last_instance_count == SI_INSTANCE_COUNT_UNKNOWN ||
+ sctx->last_instance_count != instance_count) {
+ radeon_emit(cs, PKT3(PKT3_NUM_INSTANCES, 0, 0));
+ radeon_emit(cs, instance_count);
+ sctx->last_instance_count = instance_count;
+ }
/* Base vertex and start instance. */
base_vertex = index_size ? info->index_bias : info->start;
/* Necessary for DCC */
if (sctx->chip_class == VI)
- si_gfx_write_event_eop(sctx, V_028A90_FLUSH_AND_INV_CB_DATA_TS,
- 0, EOP_DATA_SEL_DISCARD, NULL,
- 0, 0, SI_NOT_QUERY);
+ si_cp_release_mem(sctx,
+ V_028A90_FLUSH_AND_INV_CB_DATA_TS,
+ 0, EOP_DST_SEL_MEM, EOP_INT_SEL_NONE,
+ EOP_DATA_SEL_DISCARD, NULL,
+ 0, 0, SI_NOT_QUERY);
}
if (flags & SI_CONTEXT_FLUSH_AND_INV_DB)
cp_coher_cntl |= S_0085F0_DB_ACTION_ENA(1) |
va = sctx->wait_mem_scratch->gpu_address;
sctx->wait_mem_number++;
- si_gfx_write_event_eop(sctx, cb_db_event, tc_flags,
- EOP_DATA_SEL_VALUE_32BIT,
- sctx->wait_mem_scratch, va,
- sctx->wait_mem_number, SI_NOT_QUERY);
- si_gfx_wait_fence(sctx, va, sctx->wait_mem_number, 0xffffffff);
+ si_cp_release_mem(sctx, cb_db_event, tc_flags,
+ EOP_DST_SEL_MEM,
+ EOP_INT_SEL_SEND_DATA_AFTER_WR_CONFIRM,
+ EOP_DATA_SEL_VALUE_32BIT,
+ sctx->wait_mem_scratch, va,
+ sctx->wait_mem_number, SI_NOT_QUERY);
+ si_cp_wait_mem(sctx, cs, va, sctx->wait_mem_number, 0xffffffff,
+ WAIT_REG_MEM_EQUAL);
}
/* Make sure ME is idle (it executes most packets) before continuing.
unsigned skip_atom_mask)
{
unsigned num_patches = 0;
+ /* Vega10/Raven scissor bug workaround. When any context register is
+ * written (i.e. the GPU rolls the context), PA_SC_VPORT_SCISSOR
+ * registers must be written too.
+ */
+ bool handle_scissor_bug = (sctx->family == CHIP_VEGA10 || sctx->family == CHIP_RAVEN) &&
+ !si_is_atom_dirty(sctx, &sctx->atoms.s.scissors);
bool context_roll = false; /* set correctly for GFX9 only */
context_roll |= si_emit_rasterizer_prim_state(sctx);
if (sctx->tes_shader.cso)
context_roll |= si_emit_derived_tess_state(sctx, info, &num_patches);
- if (info->count_from_stream_output)
+
+ if (handle_scissor_bug &&
+ (info->count_from_stream_output ||
+ sctx->dirty_atoms & si_atoms_that_always_roll_context() ||
+ sctx->dirty_states & si_states_that_always_roll_context() ||
+ si_prim_restart_index_changed(sctx, info)))
context_roll = true;
- /* Vega10/Raven scissor bug workaround. When any context register is
- * written (i.e. the GPU rolls the context), PA_SC_VPORT_SCISSOR
- * registers must be written too.
- */
- if ((sctx->family == CHIP_VEGA10 || sctx->family == CHIP_RAVEN) &&
- (context_roll ||
- sctx->dirty_atoms & si_atoms_that_roll_context() ||
- sctx->dirty_states & si_states_that_roll_context() ||
- si_prim_restart_index_changed(sctx, info))) {
- sctx->scissors.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
- si_mark_atom_dirty(sctx, &sctx->atoms.s.scissors);
- }
+ sctx->context_roll_counter = 0;
/* Emit state atoms. */
unsigned mask = sctx->dirty_atoms & ~skip_atom_mask;
}
sctx->dirty_states = 0;
+ if (handle_scissor_bug &&
+ (context_roll || sctx->context_roll_counter)) {
+ sctx->scissors.dirty_mask = (1 << SI_MAX_VIEWPORTS) - 1;
+ sctx->atoms.s.scissors.emit(sctx);
+ }
+
/* Emit draw states. */
si_emit_vs_state(sctx, info);
si_emit_draw_registers(sctx, info, num_patches);
}
-void si_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info)
+static void si_draw_vbo(struct pipe_context *ctx, const struct pipe_draw_info *info)
{
struct si_context *sctx = (struct si_context *)ctx;
struct si_state_rasterizer *rs = sctx->queued.named.rasterizer;
pipe_resource_reference(&indexbuf, NULL);
}
-void si_draw_rectangle(struct blitter_context *blitter,
- void *vertex_elements_cso,
- blitter_get_vs_func get_vs,
- int x1, int y1, int x2, int y2,
- float depth, unsigned num_instances,
- enum blitter_attrib_type type,
- const union blitter_attrib *attrib)
+static void
+si_draw_rectangle(struct blitter_context *blitter,
+ void *vertex_elements_cso,
+ blitter_get_vs_func get_vs,
+ int x1, int y1, int x2, int y2,
+ float depth, unsigned num_instances,
+ enum blitter_attrib_type type,
+ const union blitter_attrib *attrib)
{
struct pipe_context *pipe = util_blitter_get_pipe(blitter);
struct si_context *sctx = (struct si_context*)pipe;
uint32_t trace_id = ++sctx->current_saved_cs->trace_id;
radeon_emit(cs, PKT3(PKT3_WRITE_DATA, 3, 0));
- radeon_emit(cs, S_370_DST_SEL(V_370_MEMORY_SYNC) |
+ radeon_emit(cs, S_370_DST_SEL(sctx->chip_class >= CIK ? V_370_MEM
+ : V_370_MEM_GRBM) |
S_370_WR_CONFIRM(1) |
S_370_ENGINE_SEL(V_370_ME));
radeon_emit(cs, va);
if (sctx->log)
u_log_flush(sctx->log);
}
+
+void si_init_draw_functions(struct si_context *sctx)
+{
+ sctx->b.draw_vbo = si_draw_vbo;
+
+ sctx->blitter->draw_rectangle = si_draw_rectangle;
+
+ si_init_ia_multi_vgt_param_table(sctx);
+}