(!old_cso || memcmp(old_cso->x, new_cso->x, sizeof(old_cso->x)) != 0)
static void
-flush_for_state_base_change(struct iris_batch *batch)
+flush_before_state_base_change(struct iris_batch *batch)
{
/* Flush before emitting STATE_BASE_ADDRESS.
*
* rendering. It's a bit of a big hammer but it appears to work.
*/
iris_emit_end_of_pipe_sync(batch,
- "change STATE_BASE_ADDRESS",
+ "change STATE_BASE_ADDRESS (flushes)",
PIPE_CONTROL_RENDER_TARGET_FLUSH |
PIPE_CONTROL_DEPTH_CACHE_FLUSH |
PIPE_CONTROL_DATA_CACHE_FLUSH);
}
+static void
+flush_after_state_base_change(struct iris_batch *batch)
+{
+ /* After re-setting the surface state base address, we have to do some
+ * cache flusing so that the sampler engine will pick up the new
+ * SURFACE_STATE objects and binding tables. From the Broadwell PRM,
+ * Shared Function > 3D Sampler > State > State Caching (page 96):
+ *
+ * Coherency with system memory in the state cache, like the texture
+ * cache is handled partially by software. It is expected that the
+ * command stream or shader will issue Cache Flush operation or
+ * Cache_Flush sampler message to ensure that the L1 cache remains
+ * coherent with system memory.
+ *
+ * [...]
+ *
+ * Whenever the value of the Dynamic_State_Base_Addr,
+ * Surface_State_Base_Addr are altered, the L1 state cache must be
+ * invalidated to ensure the new surface or sampler state is fetched
+ * from system memory.
+ *
+ * The PIPE_CONTROL command has a "State Cache Invalidation Enable" bit
+ * which, according the PIPE_CONTROL instruction documentation in the
+ * Broadwell PRM:
+ *
+ * Setting this bit is independent of any other bit in this packet.
+ * This bit controls the invalidation of the L1 and L2 state caches
+ * at the top of the pipe i.e. at the parsing time.
+ *
+ * Unfortunately, experimentation seems to indicate that state cache
+ * invalidation through a PIPE_CONTROL does nothing whatsoever in
+ * regards to surface state and binding tables. In stead, it seems that
+ * invalidating the texture cache is what is actually needed.
+ *
+ * XXX: As far as we have been able to determine through
+ * experimentation, shows that flush the texture cache appears to be
+ * sufficient. The theory here is that all of the sampling/rendering
+ * units cache the binding table in the texture cache. However, we have
+ * yet to be able to actually confirm this.
+ */
+ iris_emit_end_of_pipe_sync(batch,
+ "change STATE_BASE_ADDRESS (invalidates)",
+ PIPE_CONTROL_TEXTURE_CACHE_INVALIDATE |
+ PIPE_CONTROL_CONST_CACHE_INVALIDATE |
+ PIPE_CONTROL_STATE_CACHE_INVALIDATE);
+}
+
static void
_iris_emit_lri(struct iris_batch *batch, uint32_t reg, uint32_t val)
{
static void
init_state_base_address(struct iris_batch *batch)
{
- flush_for_state_base_change(batch);
+ flush_before_state_base_change(batch);
/* We program most base addresses once at context initialization time.
* Each base address points at a 4GB memory zone, and never needs to
sba.DynamicStateMOCS = MOCS_WB;
sba.IndirectObjectMOCS = MOCS_WB;
sba.InstructionMOCS = MOCS_WB;
+ sba.SurfaceStateMOCS = MOCS_WB;
sba.GeneralStateBaseAddressModifyEnable = true;
sba.DynamicStateBaseAddressModifyEnable = true;
sba.InstructionBufferSize = 0xfffff;
sba.DynamicStateBufferSize = 0xfffff;
}
+
+ flush_after_state_base_change(batch);
}
static void
bool has_slm, bool wants_dc_cache)
{
uint32_t reg_val;
- iris_pack_state(GENX(L3CNTLREG), ®_val, reg) {
+
+#if GEN_GEN >= 12
+#define L3_ALLOCATION_REG GENX(L3ALLOC)
+#define L3_ALLOCATION_REG_num GENX(L3ALLOC_num)
+#else
+#define L3_ALLOCATION_REG GENX(L3CNTLREG)
+#define L3_ALLOCATION_REG_num GENX(L3CNTLREG_num)
+#endif
+
+ iris_pack_state(L3_ALLOCATION_REG, ®_val, reg) {
+#if GEN_GEN < 12
reg.SLMEnable = has_slm;
+#endif
#if GEN_GEN == 11
/* WA_1406697149: Bit 9 "Error Detection Behavior Control" must be set
* in L3CNTLREG register. The default setting of the bit is not the
reg.DCAllocation = cfg->n[GEN_L3P_DC];
reg.AllAllocation = cfg->n[GEN_L3P_ALL];
}
- iris_emit_lri(batch, L3CNTLREG, reg_val);
+ _iris_emit_lri(batch, L3_ALLOCATION_REG_num, reg_val);
}
static void
}
iris_emit_lri(batch, HALF_SLICE_CHICKEN7, reg_val);
- iris_pack_state(GENX(SLICE_COMMON_ECO_CHICKEN1), ®_val, reg) {
- reg.StateCacheRedirectToCSSectionEnable = true;
- reg.StateCacheRedirectToCSSectionEnableMask = true;
- }
- iris_emit_lri(batch, SLICE_COMMON_ECO_CHICKEN1, reg_val);
-
/* Hardware specification recommends disabling repacking for the
* compatibility with decompression mechanism in display controller.
*/
const unsigned line_stipple_factor = state->line_stipple_factor + 1;
iris_pack_command(GENX(3DSTATE_LINE_STIPPLE), cso->line_stipple, line) {
- line.LineStipplePattern = state->line_stipple_pattern;
- line.LineStippleInverseRepeatCount = 1.0f / line_stipple_factor;
- line.LineStippleRepeatCount = line_stipple_factor;
+ if (state->line_stipple_enable) {
+ line.LineStipplePattern = state->line_stipple_pattern;
+ line.LineStippleInverseRepeatCount = 1.0f / line_stipple_factor;
+ line.LineStippleRepeatCount = line_stipple_factor;
+ }
}
return cso;
assert(start + count <= IRIS_MAX_TEXTURE_SAMPLERS);
+ bool dirty = false;
+
for (int i = 0; i < count; i++) {
- shs->samplers[start + i] = states[i];
+ if (shs->samplers[start + i] != states[i]) {
+ shs->samplers[start + i] = states[i];
+ dirty = true;
+ }
}
- ice->state.dirty |= IRIS_DIRTY_SAMPLER_STATES_VS << stage;
+ if (dirty)
+ ice->state.dirty |= IRIS_DIRTY_SAMPLER_STATES_VS << stage;
}
/**
enum isl_aux_usage aux_usage)
{
return SURFACE_STATE_ALIGNMENT *
- util_bitcount(res->aux.possible_usages & ((1 << aux_usage) - 1));
+ util_bitcount(aux_modes & ((1 << aux_usage) - 1));
}
+#if GEN_GEN == 9
static void
surf_state_update_clear_value(struct iris_batch *batch,
struct iris_resource *res,
{
struct isl_device *isl_dev = &batch->screen->isl_dev;
struct iris_bo *state_bo = iris_resource_bo(state->res);
- uint64_t real_offset = state->offset +
- IRIS_MEMZONE_BINDER_START;
+ uint64_t real_offset = state->offset + IRIS_MEMZONE_BINDER_START;
uint32_t offset_into_bo = real_offset - state_bo->gtt_offset;
uint32_t clear_offset = offset_into_bo +
isl_dev->ss.clear_value_offset +
surf_state_offset_for_aux(res, aux_modes, aux_usage);
+ uint32_t *color = res->aux.clear_color.u32;
+
+ assert(isl_dev->ss.clear_value_size == 16);
- batch->vtbl->copy_mem_mem(batch, state_bo, clear_offset,
- res->aux.clear_color_bo,
- res->aux.clear_color_offset,
- isl_dev->ss.clear_value_size);
+ if (aux_usage == ISL_AUX_USAGE_HIZ) {
+ iris_emit_pipe_control_write(batch, "update fast clear value (Z)",
+ PIPE_CONTROL_WRITE_IMMEDIATE,
+ state_bo, clear_offset, color[0]);
+ } else {
+ iris_emit_pipe_control_write(batch, "update fast clear color (RG__)",
+ PIPE_CONTROL_WRITE_IMMEDIATE,
+ state_bo, clear_offset,
+ (uint64_t) color[0] |
+ (uint64_t) color[1] << 32);
+ iris_emit_pipe_control_write(batch, "update fast clear color (__BA)",
+ PIPE_CONTROL_WRITE_IMMEDIATE,
+ state_bo, clear_offset + 8,
+ (uint64_t) color[2] |
+ (uint64_t) color[3] << 32);
+ }
+
+ iris_emit_pipe_control_flush(batch,
+ "update fast clear: state cache invalidate",
+ PIPE_CONTROL_FLUSH_ENABLE |
+ PIPE_CONTROL_STATE_CACHE_INVALIDATE);
}
+#endif
static void
update_clear_value(struct iris_context *ice,
struct iris_batch *batch,
struct iris_resource *res,
struct iris_state_ref *state,
- unsigned aux_modes,
+ unsigned all_aux_modes,
struct isl_view *view)
{
UNUSED struct isl_device *isl_dev = &batch->screen->isl_dev;
+ UNUSED unsigned aux_modes = all_aux_modes;
/* We only need to update the clear color in the surface state for gen8 and
* gen9. Newer gens can read it directly from the clear color state buffer.
while (aux_modes) {
enum isl_aux_usage aux_usage = u_bit_scan(&aux_modes);
- surf_state_update_clear_value(batch, res, state, aux_modes, aux_usage);
+ surf_state_update_clear_value(batch, res, state, all_aux_modes,
+ aux_usage);
}
#elif GEN_GEN == 8
pipe_resource_reference(&state->res, NULL);
void *map = alloc_surface_states(ice->state.surface_uploader,
- state, res->aux.possible_usages);
-
+ state, all_aux_modes);
while (aux_modes) {
enum isl_aux_usage aux_usage = u_bit_scan(&aux_modes);
fill_surface_state(isl_dev, map, res, &res->surf, view, aux_usage, 0, 0);
if (batch->last_surface_base_address == binder->bo->gtt_offset)
return;
- flush_for_state_base_change(batch);
+ flush_before_state_base_change(batch);
iris_emit_cmd(batch, GENX(STATE_BASE_ADDRESS), sba) {
- sba.SurfaceStateMOCS = MOCS_WB;
sba.SurfaceStateBaseAddressModifyEnable = true;
sba.SurfaceStateBaseAddress = ro_bo(binder->bo, 0);
+
+ /* The hardware appears to pay attention to the MOCS fields even
+ * if you don't set the "Address Modify Enable" bit for the base.
+ */
+ sba.GeneralStateMOCS = MOCS_WB;
+ sba.StatelessDataPortAccessMOCS = MOCS_WB;
+ sba.DynamicStateMOCS = MOCS_WB;
+ sba.IndirectObjectMOCS = MOCS_WB;
+ sba.InstructionMOCS = MOCS_WB;
+ sba.SurfaceStateMOCS = MOCS_WB;
+#if GEN_GEN >= 9
+ sba.BindlessSurfaceStateMOCS = MOCS_WB;
+#endif
}
+ flush_after_state_base_change(batch);
+
batch->last_surface_base_address = binder->bo->gtt_offset;
}