diff options
Diffstat (limited to 'src/gallium')
| -rw-r--r-- | src/gallium/drivers/iris/iris_blit.c | 49 | ||||
| -rw-r--r-- | src/gallium/drivers/iris/iris_resolve.c | 854 | ||||
| -rw-r--r-- | src/gallium/drivers/iris/iris_resource.c | 5 | ||||
| -rw-r--r-- | src/gallium/drivers/iris/iris_resource.h | 148 | ||||
| -rw-r--r-- | src/gallium/drivers/iris/iris_state.c | 2 |
5 files changed, 1050 insertions, 8 deletions
diff --git a/src/gallium/drivers/iris/iris_blit.c b/src/gallium/drivers/iris/iris_blit.c index d36b998bef4..b2a01df0690 100644 --- a/src/gallium/drivers/iris/iris_blit.c +++ b/src/gallium/drivers/iris/iris_blit.c @@ -239,7 +239,17 @@ iris_blorp_surf_for_resource(struct blorp_surf *surf, .aux_usage = aux_usage, }; - assert(surf->aux_usage == ISL_AUX_USAGE_NONE); + if (aux_usage != ISL_AUX_USAGE_NONE) { + surf->aux_surf = &res->aux.surf; + surf->aux_addr = (struct blorp_address) { + .buffer = res->aux.bo, + .offset = res->aux.offset, + .reloc_flags = is_render_target ? EXEC_OBJECT_WRITE : 0, + .mocs = I915_MOCS_CACHED, + }; + } + + // XXX: ASTC } /** @@ -255,6 +265,8 @@ iris_blit(struct pipe_context *ctx, const struct pipe_blit_info *info) struct iris_screen *screen = (struct iris_screen *)ctx->screen; const struct gen_device_info *devinfo = &screen->devinfo; enum blorp_batch_flags blorp_flags = 0; + struct iris_resource *src_res = (void *) info->src.resource; + struct iris_resource *dst_res = (void *) info->dst.resource; /* We don't support color masking. */ assert((info->mask & PIPE_MASK_RGBA) == PIPE_MASK_RGBA || @@ -268,18 +280,38 @@ iris_blit(struct pipe_context *ctx, const struct pipe_blit_info *info) blorp_flags |= BLORP_BATCH_PREDICATE_ENABLE; } - struct blorp_surf src_surf, dst_surf; - iris_blorp_surf_for_resource(&src_surf, info->src.resource, - ISL_AUX_USAGE_NONE, false); - iris_blorp_surf_for_resource(&dst_surf, info->dst.resource, - ISL_AUX_USAGE_NONE, true); - struct iris_format_info src_fmt = iris_format_for_usage(devinfo, info->src.format, ISL_SURF_USAGE_TEXTURE_BIT); + enum isl_aux_usage src_aux_usage = + iris_resource_texture_aux_usage(ice, src_res, src_fmt.fmt, 0); + + if (src_aux_usage == ISL_AUX_USAGE_HIZ) + src_aux_usage = ISL_AUX_USAGE_NONE; + + bool src_clear_supported = src_aux_usage != ISL_AUX_USAGE_NONE && + src_res->surf.format == src_fmt.fmt; + + iris_resource_prepare_access(ice, src_res, info->src.level, 1, + info->src.box.z, info->src.box.depth, + src_aux_usage, src_clear_supported); + struct iris_format_info dst_fmt = iris_format_for_usage(devinfo, info->dst.format, ISL_SURF_USAGE_RENDER_TARGET_BIT); + enum isl_aux_usage dst_aux_usage = + iris_resource_render_aux_usage(ice, dst_res, dst_fmt.fmt, false, false); + bool dst_clear_supported = dst_aux_usage != ISL_AUX_USAGE_NONE; + + struct blorp_surf src_surf, dst_surf; + iris_blorp_surf_for_resource(&src_surf, info->src.resource, + ISL_AUX_USAGE_NONE, false); + iris_blorp_surf_for_resource(&dst_surf, info->dst.resource, + ISL_AUX_USAGE_NONE, true); + + iris_resource_prepare_access(ice, dst_res, info->dst.level, 1, + info->dst.box.z, info->dst.box.depth, + dst_aux_usage, dst_clear_supported); float src_x0 = info->src.box.x; float src_x1 = info->src.box.x + info->src.box.width; @@ -403,6 +435,9 @@ iris_blit(struct pipe_context *ctx, const struct pipe_blit_info *info) blorp_batch_finish(&blorp_batch); + iris_resource_finish_write(ice, dst_res, info->dst.level, info->dst.box.z, + info->dst.box.depth, dst_aux_usage); + iris_flush_and_dirty_for_history(ice, batch, (struct iris_resource *) info->dst.resource); } diff --git a/src/gallium/drivers/iris/iris_resolve.c b/src/gallium/drivers/iris/iris_resolve.c index 737bcfe0e90..9fc32fc0482 100644 --- a/src/gallium/drivers/iris/iris_resolve.c +++ b/src/gallium/drivers/iris/iris_resolve.c @@ -286,3 +286,857 @@ iris_depth_cache_add_bo(struct iris_batch *batch, struct iris_bo *bo) { _mesa_set_add_pre_hashed(batch->cache.depth, bo->hash, bo); } + +/** + * Return true if the format that will be used to access the resource is + * CCS_E-compatible with the resource's linear/non-sRGB format. + * + * Why use the linear format? Well, although the resourcemay be specified + * with an sRGB format, the usage of that color space/format can be toggled. + * Since our HW tends to support more linear formats than sRGB ones, we use + * this format variant for check for CCS_E compatibility. + */ +static bool +format_ccs_e_compat_with_resource(const struct gen_device_info *devinfo, + const struct iris_resource *res, + enum isl_format access_format) +{ + assert(res->aux.usage == ISL_AUX_USAGE_CCS_E); + + enum isl_format isl_format = isl_format_srgb_to_linear(res->surf.format); + return isl_formats_are_ccs_e_compatible(devinfo, isl_format, access_format); +} + +static bool +sample_with_hiz(const struct gen_device_info *devinfo, + const struct iris_resource *res) +{ + if (!devinfo->has_sample_with_hiz) + return false; + + if (res->aux.usage != ISL_AUX_USAGE_HIZ) + return false; + + /* It seems the hardware won't fallback to the depth buffer if some of the + * mipmap levels aren't available in the HiZ buffer. So we need all levels + * of the texture to be HiZ enabled. + */ + for (unsigned level = 0; level < res->surf.levels; ++level) { + if (!iris_resource_level_has_hiz(res, level)) + return false; + } + + /* If compressed multisampling is enabled, then we use it for the auxiliary + * buffer instead. + * + * From the BDW PRM (Volume 2d: Command Reference: Structures + * RENDER_SURFACE_STATE.AuxiliarySurfaceMode): + * + * "If this field is set to AUX_HIZ, Number of Multisamples must be + * MULTISAMPLECOUNT_1, and Surface Type cannot be SURFTYPE_3D. + * + * There is no such blurb for 1D textures, but there is sufficient evidence + * that this is broken on SKL+. + */ + // XXX: i965 disables this for arrays too, is that reasonable? + return res->surf.samples == 1 && res->surf.dim == ISL_SURF_DIM_2D; +} + +/** + * Does the resource's slice have hiz enabled? + */ +bool +iris_resource_level_has_hiz(const struct iris_resource *res, uint32_t level) +{ + iris_resource_check_level_layer(res, level, 0); + // return res->level[level].has_hiz; + return false; +} + +/** \brief Assert that the level and layer are valid for the resource. */ +void +iris_resource_check_level_layer(UNUSED const struct iris_resource *res, + UNUSED uint32_t level, UNUSED uint32_t layer) +{ + assert(level < res->surf.levels); + assert(layer < util_num_layers(&res->base, level)); +} + +static inline uint32_t +miptree_level_range_length(const struct iris_resource *res, + uint32_t start_level, uint32_t num_levels) +{ + assert(start_level < res->surf.levels); + + if (num_levels == INTEL_REMAINING_LAYERS) + num_levels = res->surf.levels; + + /* Check for overflow */ + assert(start_level + num_levels >= start_level); + assert(start_level + num_levels <= res->surf.levels); + + return num_levels; +} + +static inline uint32_t +miptree_layer_range_length(const struct iris_resource *res, uint32_t level, + uint32_t start_layer, uint32_t num_layers) +{ + assert(level <= res->base.last_level); + + const uint32_t total_num_layers = iris_get_num_logical_layers(res, level); + assert(start_layer < total_num_layers); + if (num_layers == INTEL_REMAINING_LAYERS) + num_layers = total_num_layers - start_layer; + /* Check for overflow */ + assert(start_layer + num_layers >= start_layer); + assert(start_layer + num_layers <= total_num_layers); + + return num_layers; +} + +static bool +has_color_unresolved(const struct iris_resource *res, + unsigned start_level, unsigned num_levels, + unsigned start_layer, unsigned num_layers) +{ + if (!res->aux.bo) + return false; + + /* Clamp the level range to fit the resource */ + num_levels = miptree_level_range_length(res, start_level, num_levels); + + for (uint32_t l = 0; l < num_levels; l++) { + const uint32_t level = start_level + l; + const uint32_t level_layers = + miptree_layer_range_length(res, level, start_layer, num_layers); + for (unsigned a = 0; a < level_layers; a++) { + enum isl_aux_state aux_state = + iris_resource_get_aux_state(res, level, start_layer + a); + assert(aux_state != ISL_AUX_STATE_AUX_INVALID); + if (aux_state != ISL_AUX_STATE_PASS_THROUGH) + return true; + } + } + + return false; +} + +static enum isl_aux_op +get_ccs_d_resolve_op(enum isl_aux_state aux_state, + enum isl_aux_usage aux_usage, + bool fast_clear_supported) +{ + assert(aux_usage == ISL_AUX_USAGE_NONE || aux_usage == ISL_AUX_USAGE_CCS_D); + + const bool ccs_supported = aux_usage == ISL_AUX_USAGE_CCS_D; + + assert(ccs_supported == fast_clear_supported); + + switch (aux_state) { + case ISL_AUX_STATE_CLEAR: + case ISL_AUX_STATE_PARTIAL_CLEAR: + if (!ccs_supported) + return ISL_AUX_OP_FULL_RESOLVE; + else + return ISL_AUX_OP_NONE; + + case ISL_AUX_STATE_PASS_THROUGH: + return ISL_AUX_OP_NONE; + + case ISL_AUX_STATE_RESOLVED: + case ISL_AUX_STATE_AUX_INVALID: + case ISL_AUX_STATE_COMPRESSED_CLEAR: + case ISL_AUX_STATE_COMPRESSED_NO_CLEAR: + break; + } + + unreachable("Invalid aux state for CCS_D"); +} + +static enum isl_aux_op +get_ccs_e_resolve_op(enum isl_aux_state aux_state, + enum isl_aux_usage aux_usage, + bool fast_clear_supported) +{ + /* CCS_E surfaces can be accessed as CCS_D if we're careful. */ + assert(aux_usage == ISL_AUX_USAGE_NONE || + aux_usage == ISL_AUX_USAGE_CCS_D || + aux_usage == ISL_AUX_USAGE_CCS_E); + + if (aux_usage == ISL_AUX_USAGE_CCS_D) + assert(fast_clear_supported); + + switch (aux_state) { + case ISL_AUX_STATE_CLEAR: + case ISL_AUX_STATE_PARTIAL_CLEAR: + if (fast_clear_supported) + return ISL_AUX_OP_NONE; + else if (aux_usage == ISL_AUX_USAGE_CCS_E) + return ISL_AUX_OP_PARTIAL_RESOLVE; + else + return ISL_AUX_OP_FULL_RESOLVE; + + case ISL_AUX_STATE_COMPRESSED_CLEAR: + if (aux_usage != ISL_AUX_USAGE_CCS_E) + return ISL_AUX_OP_FULL_RESOLVE; + else if (!fast_clear_supported) + return ISL_AUX_OP_PARTIAL_RESOLVE; + else + return ISL_AUX_OP_NONE; + + case ISL_AUX_STATE_COMPRESSED_NO_CLEAR: + if (aux_usage != ISL_AUX_USAGE_CCS_E) + return ISL_AUX_OP_FULL_RESOLVE; + else + return ISL_AUX_OP_NONE; + + case ISL_AUX_STATE_PASS_THROUGH: + return ISL_AUX_OP_NONE; + + case ISL_AUX_STATE_RESOLVED: + case ISL_AUX_STATE_AUX_INVALID: + break; + } + + unreachable("Invalid aux state for CCS_E"); +} + +static void +iris_resource_prepare_ccs_access(struct iris_context *ice, + struct iris_resource *res, + uint32_t level, uint32_t layer, + enum isl_aux_usage aux_usage, + bool fast_clear_supported) +{ + enum isl_aux_state aux_state = iris_resource_get_aux_state(res, level, layer); + + enum isl_aux_op resolve_op; + if (res->aux.usage == ISL_AUX_USAGE_CCS_E) { + resolve_op = get_ccs_e_resolve_op(aux_state, aux_usage, + fast_clear_supported); + } else { + assert(res->aux.usage == ISL_AUX_USAGE_CCS_D); + resolve_op = get_ccs_d_resolve_op(aux_state, aux_usage, + fast_clear_supported); + } + + if (resolve_op != ISL_AUX_OP_NONE) { + // XXX: iris_blorp_resolve_color(ice, res, level, layer, resolve_op); + + switch (resolve_op) { + case ISL_AUX_OP_FULL_RESOLVE: + /* The CCS full resolve operation destroys the CCS and sets it to the + * pass-through state. (You can also think of this as being both a + * resolve and an ambiguate in one operation.) + */ + iris_resource_set_aux_state(res, level, layer, 1, + ISL_AUX_STATE_PASS_THROUGH); + break; + + case ISL_AUX_OP_PARTIAL_RESOLVE: + iris_resource_set_aux_state(res, level, layer, 1, + ISL_AUX_STATE_COMPRESSED_NO_CLEAR); + break; + + default: + unreachable("Invalid resolve op"); + } + } +} + +static void +iris_resource_finish_ccs_write(struct iris_context *ice, + struct iris_resource *res, + uint32_t level, uint32_t layer, + enum isl_aux_usage aux_usage) +{ + assert(aux_usage == ISL_AUX_USAGE_NONE || + aux_usage == ISL_AUX_USAGE_CCS_D || + aux_usage == ISL_AUX_USAGE_CCS_E); + + enum isl_aux_state aux_state = + iris_resource_get_aux_state(res, level, layer); + + if (res->aux.usage == ISL_AUX_USAGE_CCS_E) { + switch (aux_state) { + case ISL_AUX_STATE_CLEAR: + case ISL_AUX_STATE_PARTIAL_CLEAR: + assert(aux_usage == ISL_AUX_USAGE_CCS_E || + aux_usage == ISL_AUX_USAGE_CCS_D); + + if (aux_usage == ISL_AUX_USAGE_CCS_E) { + iris_resource_set_aux_state(res, level, layer, 1, + ISL_AUX_STATE_COMPRESSED_CLEAR); + } else if (aux_state != ISL_AUX_STATE_PARTIAL_CLEAR) { + iris_resource_set_aux_state(res, level, layer, 1, + ISL_AUX_STATE_PARTIAL_CLEAR); + } + break; + + case ISL_AUX_STATE_COMPRESSED_CLEAR: + case ISL_AUX_STATE_COMPRESSED_NO_CLEAR: + assert(aux_usage == ISL_AUX_USAGE_CCS_E); + break; /* Nothing to do */ + + case ISL_AUX_STATE_PASS_THROUGH: + if (aux_usage == ISL_AUX_USAGE_CCS_E) { + iris_resource_set_aux_state(res, level, layer, 1, + ISL_AUX_STATE_COMPRESSED_NO_CLEAR); + } else { + /* Nothing to do */ + } + break; + + case ISL_AUX_STATE_RESOLVED: + case ISL_AUX_STATE_AUX_INVALID: + unreachable("Invalid aux state for CCS_E"); + } + } else { + assert(res->aux.usage == ISL_AUX_USAGE_CCS_D); + /* CCS_D is a bit simpler */ + switch (aux_state) { + case ISL_AUX_STATE_CLEAR: + assert(aux_usage == ISL_AUX_USAGE_CCS_D); + iris_resource_set_aux_state(res, level, layer, 1, + ISL_AUX_STATE_PARTIAL_CLEAR); + break; + + case ISL_AUX_STATE_PARTIAL_CLEAR: + assert(aux_usage == ISL_AUX_USAGE_CCS_D); + break; /* Nothing to do */ + + case ISL_AUX_STATE_PASS_THROUGH: + /* Nothing to do */ + break; + + case ISL_AUX_STATE_COMPRESSED_CLEAR: + case ISL_AUX_STATE_COMPRESSED_NO_CLEAR: + case ISL_AUX_STATE_RESOLVED: + case ISL_AUX_STATE_AUX_INVALID: + unreachable("Invalid aux state for CCS_D"); + } + } +} + +static void +iris_resource_prepare_mcs_access(struct iris_context *ice, + struct iris_resource *res, + uint32_t layer, + enum isl_aux_usage aux_usage, + bool fast_clear_supported) +{ + assert(aux_usage == ISL_AUX_USAGE_MCS); + + switch (iris_resource_get_aux_state(res, 0, layer)) { + case ISL_AUX_STATE_CLEAR: + case ISL_AUX_STATE_COMPRESSED_CLEAR: + if (!fast_clear_supported) { + // XXX: iris_blorp_mcs_partial_resolve(ice, res, layer, 1); + iris_resource_set_aux_state(res, 0, layer, 1, + ISL_AUX_STATE_COMPRESSED_NO_CLEAR); + } + break; + + case ISL_AUX_STATE_COMPRESSED_NO_CLEAR: + break; /* Nothing to do */ + + case ISL_AUX_STATE_RESOLVED: + case ISL_AUX_STATE_PASS_THROUGH: + case ISL_AUX_STATE_AUX_INVALID: + case ISL_AUX_STATE_PARTIAL_CLEAR: + unreachable("Invalid aux state for MCS"); + } +} + +static void +iris_resource_finish_mcs_write(struct iris_context *ice, + struct iris_resource *res, + uint32_t layer, + enum isl_aux_usage aux_usage) +{ + assert(aux_usage == ISL_AUX_USAGE_MCS); + + switch (iris_resource_get_aux_state(res, 0, layer)) { + case ISL_AUX_STATE_CLEAR: + iris_resource_set_aux_state(res, 0, layer, 1, + ISL_AUX_STATE_COMPRESSED_CLEAR); + break; + + case ISL_AUX_STATE_COMPRESSED_CLEAR: + case ISL_AUX_STATE_COMPRESSED_NO_CLEAR: + break; /* Nothing to do */ + + case ISL_AUX_STATE_RESOLVED: + case ISL_AUX_STATE_PASS_THROUGH: + case ISL_AUX_STATE_AUX_INVALID: + case ISL_AUX_STATE_PARTIAL_CLEAR: + unreachable("Invalid aux state for MCS"); + } +} + +static void +iris_resource_prepare_hiz_access(struct iris_context *ice, + struct iris_resource *res, + uint32_t level, uint32_t layer, + enum isl_aux_usage aux_usage, + bool fast_clear_supported) +{ + assert(aux_usage == ISL_AUX_USAGE_NONE || aux_usage == ISL_AUX_USAGE_HIZ); + + enum isl_aux_op hiz_op = ISL_AUX_OP_NONE; + switch (iris_resource_get_aux_state(res, level, layer)) { + case ISL_AUX_STATE_CLEAR: + case ISL_AUX_STATE_COMPRESSED_CLEAR: + if (aux_usage != ISL_AUX_USAGE_HIZ || !fast_clear_supported) + hiz_op = ISL_AUX_OP_FULL_RESOLVE; + break; + + case ISL_AUX_STATE_COMPRESSED_NO_CLEAR: + if (aux_usage != ISL_AUX_USAGE_HIZ) + hiz_op = ISL_AUX_OP_FULL_RESOLVE; + break; + + case ISL_AUX_STATE_PASS_THROUGH: + case ISL_AUX_STATE_RESOLVED: + break; + + case ISL_AUX_STATE_AUX_INVALID: + if (aux_usage == ISL_AUX_USAGE_HIZ) + hiz_op = ISL_AUX_OP_AMBIGUATE; + break; + + case ISL_AUX_STATE_PARTIAL_CLEAR: + unreachable("Invalid HiZ state"); + } + + if (hiz_op != ISL_AUX_OP_NONE) { + // XXX: HiZ + //intel_hiz_exec(ice, res, level, layer, 1, hiz_op); + + switch (hiz_op) { + case ISL_AUX_OP_FULL_RESOLVE: + iris_resource_set_aux_state(res, level, layer, 1, + ISL_AUX_STATE_RESOLVED); + break; + + case ISL_AUX_OP_AMBIGUATE: + /* The HiZ resolve operation is actually an ambiguate */ + iris_resource_set_aux_state(res, level, layer, 1, + ISL_AUX_STATE_PASS_THROUGH); + break; + + default: + unreachable("Invalid HiZ op"); + } + } +} + +static void +iris_resource_finish_hiz_write(struct iris_context *ice, + struct iris_resource *res, + uint32_t level, uint32_t layer, + enum isl_aux_usage aux_usage) +{ + assert(aux_usage == ISL_AUX_USAGE_NONE || aux_usage == ISL_AUX_USAGE_HIZ); + + switch (iris_resource_get_aux_state(res, level, layer)) { + case ISL_AUX_STATE_CLEAR: + assert(aux_usage == ISL_AUX_USAGE_HIZ); + iris_resource_set_aux_state(res, level, layer, 1, + ISL_AUX_STATE_COMPRESSED_CLEAR); + break; + + case ISL_AUX_STATE_COMPRESSED_NO_CLEAR: + case ISL_AUX_STATE_COMPRESSED_CLEAR: + assert(aux_usage == ISL_AUX_USAGE_HIZ); + break; /* Nothing to do */ + + case ISL_AUX_STATE_RESOLVED: + if (aux_usage == ISL_AUX_USAGE_HIZ) { + iris_resource_set_aux_state(res, level, layer, 1, + ISL_AUX_STATE_COMPRESSED_NO_CLEAR); + } else { + iris_resource_set_aux_state(res, level, layer, 1, + ISL_AUX_STATE_AUX_INVALID); + } + break; + + case ISL_AUX_STATE_PASS_THROUGH: + if (aux_usage == ISL_AUX_USAGE_HIZ) { + iris_resource_set_aux_state(res, level, layer, 1, + ISL_AUX_STATE_COMPRESSED_NO_CLEAR); + } + break; + + case ISL_AUX_STATE_AUX_INVALID: + assert(aux_usage != ISL_AUX_USAGE_HIZ); + break; + + case ISL_AUX_STATE_PARTIAL_CLEAR: + unreachable("Invalid HiZ state"); + } +} + +void +iris_resource_prepare_access(struct iris_context *ice, + struct iris_resource *res, + uint32_t start_level, uint32_t num_levels, + uint32_t start_layer, uint32_t num_layers, + enum isl_aux_usage aux_usage, + bool fast_clear_supported) +{ + num_levels = miptree_level_range_length(res, start_level, num_levels); + + switch (res->aux.usage) { + case ISL_AUX_USAGE_NONE: + /* Nothing to do */ + break; + + case ISL_AUX_USAGE_MCS: + assert(start_level == 0 && num_levels == 1); + const uint32_t level_layers = + miptree_layer_range_length(res, 0, start_layer, num_layers); + for (uint32_t a = 0; a < level_layers; a++) { + iris_resource_prepare_mcs_access(ice, res, start_layer + a, + aux_usage, fast_clear_supported); + } + break; + + case ISL_AUX_USAGE_CCS_D: + case ISL_AUX_USAGE_CCS_E: + for (uint32_t l = 0; l < num_levels; l++) { + const uint32_t level = start_level + l; + const uint32_t level_layers = + miptree_layer_range_length(res, level, start_layer, num_layers); + for (uint32_t a = 0; a < level_layers; a++) { + iris_resource_prepare_ccs_access(ice, res, level, + start_layer + a, + aux_usage, fast_clear_supported); + } + } + break; + + case ISL_AUX_USAGE_HIZ: + for (uint32_t l = 0; l < num_levels; l++) { + const uint32_t level = start_level + l; + if (!iris_resource_level_has_hiz(res, level)) + continue; + + const uint32_t level_layers = + miptree_layer_range_length(res, level, start_layer, num_layers); + for (uint32_t a = 0; a < level_layers; a++) { + iris_resource_prepare_hiz_access(ice, res, level, start_layer + a, + aux_usage, fast_clear_supported); + } + } + break; + + default: + unreachable("Invalid aux usage"); + } +} + +void +iris_resource_finish_write(struct iris_context *ice, + struct iris_resource *res, uint32_t level, + uint32_t start_layer, uint32_t num_layers, + enum isl_aux_usage aux_usage) +{ + num_layers = miptree_layer_range_length(res, level, start_layer, num_layers); + + switch (res->aux.usage) { + case ISL_AUX_USAGE_NONE: + break; + + case ISL_AUX_USAGE_MCS: + for (uint32_t a = 0; a < num_layers; a++) { + iris_resource_finish_mcs_write(ice, res, start_layer + a, + aux_usage); + } + break; + + case ISL_AUX_USAGE_CCS_D: + case ISL_AUX_USAGE_CCS_E: + for (uint32_t a = 0; a < num_layers; a++) { + iris_resource_finish_ccs_write(ice, res, level, start_layer + a, + aux_usage); + } + break; + + case ISL_AUX_USAGE_HIZ: + if (!iris_resource_level_has_hiz(res, level)) + return; + + for (uint32_t a = 0; a < num_layers; a++) { + iris_resource_finish_hiz_write(ice, res, level, start_layer + a, + aux_usage); + } + break; + + default: + unreachable("Invavlid aux usage"); + } +} + +enum isl_aux_state +iris_resource_get_aux_state(const struct iris_resource *res, + uint32_t level, uint32_t layer) +{ + iris_resource_check_level_layer(res, level, layer); + + if (res->surf.usage & ISL_SURF_USAGE_DEPTH_BIT) { + assert(iris_resource_level_has_hiz(res, level)); + } else if (res->surf.usage & ISL_SURF_USAGE_STENCIL_BIT) { + unreachable("Cannot get aux state for stencil"); + } else { + assert(res->surf.samples == 1 || + res->surf.msaa_layout == ISL_MSAA_LAYOUT_ARRAY); + } + + return res->aux.state[level][layer]; +} + +void +iris_resource_set_aux_state(struct iris_resource *res, uint32_t level, + uint32_t start_layer, uint32_t num_layers, + enum isl_aux_state aux_state) +{ + num_layers = miptree_layer_range_length(res, level, start_layer, num_layers); + + if (res->surf.usage & ISL_SURF_USAGE_DEPTH_BIT) { + assert(iris_resource_level_has_hiz(res, level)); + } else if (res->surf.usage & ISL_SURF_USAGE_STENCIL_BIT) { + unreachable("Cannot set aux state for stencil"); + } else { + assert(res->surf.samples == 1 || + res->surf.msaa_layout == ISL_MSAA_LAYOUT_ARRAY); + } + + for (unsigned a = 0; a < num_layers; a++) { + if (res->aux.state[level][start_layer + a] != aux_state) { + res->aux.state[level][start_layer + a] = aux_state; + // XXX: dirty works differently + // brw->ctx.NewDriverState |= BRW_NEW_AUX_STATE; + } + } +} + +/* On Gen9 color buffers may be compressed by the hardware (lossless + * compression). There are, however, format restrictions and care needs to be + * taken that the sampler engine is capable for re-interpreting a buffer with + * format different the buffer was originally written with. + * + * For example, SRGB formats are not compressible and the sampler engine isn't + * capable of treating RGBA_UNORM as SRGB_ALPHA. In such a case the underlying + * color buffer needs to be resolved so that the sampling surface can be + * sampled as non-compressed (i.e., without the auxiliary MCS buffer being + * set). + */ +static bool +can_texture_with_ccs(const struct gen_device_info *devinfo, + struct pipe_debug_callback *dbg, + const struct iris_resource *res, + enum isl_format view_format) +{ + if (res->aux.usage != ISL_AUX_USAGE_CCS_E) + return false; + + if (!format_ccs_e_compat_with_resource(devinfo, res, view_format)) { + const struct isl_format_layout *res_fmtl = + isl_format_get_layout(res->surf.format); + const struct isl_format_layout *view_fmtl = + isl_format_get_layout(view_format); + + perf_debug(dbg, "Incompatible sampling format (%s) for CCS (%s)\n", + view_fmtl->name, res_fmtl->name); + + return false; + } + + return true; +} + +enum isl_aux_usage +iris_resource_texture_aux_usage(struct iris_context *ice, + const struct iris_resource *res, + enum isl_format view_format, + enum gen9_astc5x5_wa_tex_type astc5x5_wa_bits) +{ + struct iris_screen *screen = (void *) ice->ctx.screen; + struct gen_device_info *devinfo = &screen->devinfo; + + assert(devinfo->gen == 9 || astc5x5_wa_bits == 0); + + /* On gen9, ASTC 5x5 textures cannot live in the sampler cache along side + * CCS or HiZ compressed textures. See gen9_apply_astc5x5_wa_flush() for + * details. + */ + if ((astc5x5_wa_bits & GEN9_ASTC5X5_WA_TEX_TYPE_ASTC5x5) && + res->aux.usage != ISL_AUX_USAGE_MCS) + return ISL_AUX_USAGE_NONE; + + switch (res->aux.usage) { + case ISL_AUX_USAGE_HIZ: + if (sample_with_hiz(devinfo, res)) + return ISL_AUX_USAGE_HIZ; + break; + + case ISL_AUX_USAGE_MCS: + return ISL_AUX_USAGE_MCS; + + case ISL_AUX_USAGE_CCS_D: + case ISL_AUX_USAGE_CCS_E: + /* If we don't have any unresolved color, report an aux usage of + * ISL_AUX_USAGE_NONE. This way, texturing won't even look at the + * aux surface and we can save some bandwidth. + */ + if (!has_color_unresolved(res, 0, INTEL_REMAINING_LEVELS, + 0, INTEL_REMAINING_LAYERS)) + return ISL_AUX_USAGE_NONE; + + if (can_texture_with_ccs(devinfo, &ice->dbg, res, view_format)) + return ISL_AUX_USAGE_CCS_E; + break; + + default: + break; + } + + return ISL_AUX_USAGE_NONE; +} + +static bool +isl_formats_are_fast_clear_compatible(enum isl_format a, enum isl_format b) +{ + /* On gen8 and earlier, the hardware was only capable of handling 0/1 clear + * values so sRGB curve application was a no-op for all fast-clearable + * formats. + * + * On gen9+, the hardware supports arbitrary clear values. For sRGB clear + * values, the hardware interprets the floats, not as what would be + * returned from the sampler (or written by the shader), but as being + * between format conversion and sRGB curve application. This means that + * we can switch between sRGB and UNORM without having to whack the clear + * color. + */ + return isl_format_srgb_to_linear(a) == isl_format_srgb_to_linear(b); +} + +void +iris_resource_prepare_texture(struct iris_context *ice, + struct iris_resource *res, + enum isl_format view_format, + uint32_t start_level, uint32_t num_levels, + uint32_t start_layer, uint32_t num_layers, + enum gen9_astc5x5_wa_tex_type astc5x5_wa_bits) +{ + enum isl_aux_usage aux_usage = + iris_resource_texture_aux_usage(ice, res, view_format, astc5x5_wa_bits); + + bool clear_supported = aux_usage != ISL_AUX_USAGE_NONE; + + /* Clear color is specified as ints or floats and the conversion is done by + * the sampler. If we have a texture view, we would have to perform the + * clear color conversion manually. Just disable clear color. + */ + if (!isl_formats_are_fast_clear_compatible(res->surf.format, view_format)) + clear_supported = false; + + iris_resource_prepare_access(ice, res, start_level, num_levels, + start_layer, num_layers, + aux_usage, clear_supported); +} + +void +iris_resource_prepare_image(struct iris_context *ice, + struct iris_resource *res) +{ + /* The data port doesn't understand any compression */ + iris_resource_prepare_access(ice, res, 0, INTEL_REMAINING_LEVELS, + 0, INTEL_REMAINING_LAYERS, + ISL_AUX_USAGE_NONE, false); +} + +enum isl_aux_usage +iris_resource_render_aux_usage(struct iris_context *ice, + struct iris_resource *res, + enum isl_format render_format, + bool blend_enabled, + bool draw_aux_disabled) +{ + struct iris_screen *screen = (void *) ice->ctx.screen; + struct gen_device_info *devinfo = &screen->devinfo; + + if (draw_aux_disabled) + return ISL_AUX_USAGE_NONE; + + switch (res->aux.usage) { + case ISL_AUX_USAGE_MCS: + return ISL_AUX_USAGE_MCS; + + case ISL_AUX_USAGE_CCS_D: + case ISL_AUX_USAGE_CCS_E: + /* Gen9+ hardware technically supports non-0/1 clear colors with sRGB + * formats. However, there are issues with blending where it doesn't + * properly apply the sRGB curve to the clear color when blending. + */ + /* XXX: + if (devinfo->gen >= 9 && blend_enabled && + isl_format_is_srgb(render_format) && + !isl_color_value_is_zero_one(res->fast_clear_color, render_format)) + return ISL_AUX_USAGE_NONE; + */ + + if (res->aux.usage == ISL_AUX_USAGE_CCS_E && + format_ccs_e_compat_with_resource(devinfo, res, render_format)) + return ISL_AUX_USAGE_CCS_E; + + /* Otherwise, we have to fall back to CCS_D */ + return ISL_AUX_USAGE_CCS_D; + + default: + return ISL_AUX_USAGE_NONE; + } +} + +void +iris_resource_prepare_render(struct iris_context *ice, + struct iris_resource *res, uint32_t level, + uint32_t start_layer, uint32_t layer_count, + enum isl_aux_usage aux_usage) +{ + iris_resource_prepare_access(ice, res, level, 1, start_layer, layer_count, + aux_usage, aux_usage != ISL_AUX_USAGE_NONE); +} + +void +iris_resource_finish_render(struct iris_context *ice, + struct iris_resource *res, uint32_t level, + uint32_t start_layer, uint32_t layer_count, + enum isl_aux_usage aux_usage) +{ + iris_resource_finish_write(ice, res, level, start_layer, layer_count, + aux_usage); +} + +void +iris_resource_prepare_depth(struct iris_context *ice, + struct iris_resource *res, uint32_t level, + uint32_t start_layer, uint32_t layer_count) +{ + iris_resource_prepare_access(ice, res, level, 1, start_layer, layer_count, + res->aux.usage, res->aux.bo != NULL); +} + +void +iris_resource_finish_depth(struct iris_context *ice, + struct iris_resource *res, uint32_t level, + uint32_t start_layer, uint32_t layer_count, + bool depth_written) +{ + if (depth_written) { + iris_resource_finish_write(ice, res, level, start_layer, layer_count, + res->aux.usage); + } +} diff --git a/src/gallium/drivers/iris/iris_resource.c b/src/gallium/drivers/iris/iris_resource.c index b474093956d..349e4efe022 100644 --- a/src/gallium/drivers/iris/iris_resource.c +++ b/src/gallium/drivers/iris/iris_resource.c @@ -196,12 +196,15 @@ iris_get_depth_stencil_resources(struct pipe_resource *res, } } -static void +void iris_resource_disable_aux(struct iris_resource *res) { iris_bo_unreference(res->aux.bo); free(res->aux.state); + // XXX: clear color BO + // XXX: HiZ + res->aux.usage = ISL_AUX_USAGE_NONE; res->aux.surf.size_B = 0; res->aux.bo = NULL; diff --git a/src/gallium/drivers/iris/iris_resource.h b/src/gallium/drivers/iris/iris_resource.h index 0424ea253e8..c17ed42ada2 100644 --- a/src/gallium/drivers/iris/iris_resource.h +++ b/src/gallium/drivers/iris/iris_resource.h @@ -41,6 +41,11 @@ struct iris_format_info { #define IRIS_RESOURCE_FLAG_SURFACE_MEMZONE (PIPE_RESOURCE_FLAG_DRV_PRIV << 1) #define IRIS_RESOURCE_FLAG_DYNAMIC_MEMZONE (PIPE_RESOURCE_FLAG_DRV_PRIV << 2) +enum gen9_astc5x5_wa_tex_type { + GEN9_ASTC5X5_WA_TEX_TYPE_ASTC5x5 = 1 << 0, + GEN9_ASTC5X5_WA_TEX_TYPE_AUX = 1 << 1, +}; + /** * Resources represent a GPU buffer object or image (mipmap tree). * @@ -195,4 +200,147 @@ void iris_flush_and_dirty_for_history(struct iris_context *ice, unsigned iris_get_num_logical_layers(const struct iris_resource *res, unsigned level); +void iris_resource_disable_aux(struct iris_resource *res); + +#define INTEL_REMAINING_LAYERS UINT32_MAX +#define INTEL_REMAINING_LEVELS UINT32_MAX + +/** + * Prepare a miptree for access + * + * This function should be called prior to any access to miptree in order to + * perform any needed resolves. + * + * \param[in] start_level The first mip level to be accessed + * + * \param[in] num_levels The number of miplevels to be accessed or + * INTEL_REMAINING_LEVELS to indicate every level + * above start_level will be accessed + * + * \param[in] start_layer The first array slice or 3D layer to be accessed + * + * \param[in] num_layers The number of array slices or 3D layers be + * accessed or INTEL_REMAINING_LAYERS to indicate + * every layer above start_layer will be accessed + * + * \param[in] aux_supported Whether or not the access will support the + * miptree's auxiliary compression format; this + * must be false for uncompressed miptrees + * + * \param[in] fast_clear_supported Whether or not the access will support + * fast clears in the miptree's auxiliary + * compression format + */ +void +iris_resource_prepare_access(struct iris_context *ice, + struct iris_resource *res, + uint32_t start_level, uint32_t num_levels, + uint32_t start_layer, uint32_t num_layers, + enum isl_aux_usage aux_usage, + bool fast_clear_supported); + +/** + * Complete a write operation + * + * This function should be called after any operation writes to a miptree. + * This will update the miptree's compression state so that future resolves + * happen correctly. Technically, this function can be called before the + * write occurs but the caller must ensure that they don't interlace + * iris_resource_prepare_access and iris_resource_finish_write calls to + * overlapping layer/level ranges. + * + * \param[in] level The mip level that was written + * + * \param[in] start_layer The first array slice or 3D layer written + * + * \param[in] num_layers The number of array slices or 3D layers + * written or INTEL_REMAINING_LAYERS to indicate + * every layer above start_layer was written + * + * \param[in] written_with_aux Whether or not the write was done with + * auxiliary compression enabled + */ +void +iris_resource_finish_write(struct iris_context *ice, + struct iris_resource *res, uint32_t level, + uint32_t start_layer, uint32_t num_layers, + enum isl_aux_usage aux_usage); + +/** Get the auxiliary compression state of a miptree slice */ +enum isl_aux_state +iris_resource_get_aux_state(const struct iris_resource *res, + uint32_t level, uint32_t layer); + +/** + * Set the auxiliary compression state of a miptree slice range + * + * This function directly sets the auxiliary compression state of a slice + * range of a miptree. It only modifies data structures and does not do any + * resolves. This should only be called by code which directly performs + * compression operations such as fast clears and resolves. Most code should + * use iris_resource_prepare_access or iris_resource_finish_write. + */ +void +iris_resource_set_aux_state(struct iris_resource *res, uint32_t level, + uint32_t start_layer, uint32_t num_layers, + enum isl_aux_state aux_state); + +/** + * Prepare a miptree for raw access + * + * This helper prepares the miptree for access that knows nothing about any + * sort of compression whatsoever. This is useful when mapping the surface or + * using it with the blitter. + */ +static inline void +iris_resource_access_raw(struct iris_context *ice, + struct iris_resource *res, + uint32_t level, uint32_t layer, + bool write) +{ + iris_resource_prepare_access(ice, res, level, 1, layer, 1, + ISL_AUX_USAGE_NONE, false); + if (write) + iris_resource_finish_write(ice, res, level, layer, 1, ISL_AUX_USAGE_NONE); +} + +enum isl_aux_usage iris_resource_texture_aux_usage(struct iris_context *ice, + const struct iris_resource *res, + enum isl_format view_fmt, + enum gen9_astc5x5_wa_tex_type); +void iris_resource_prepare_texture(struct iris_context *ice, + struct iris_resource *res, + enum isl_format view_format, + uint32_t start_level, uint32_t num_levels, + uint32_t start_layer, uint32_t num_layers, + enum gen9_astc5x5_wa_tex_type); +void iris_resource_prepare_image(struct iris_context *ice, + struct iris_resource *res); + +void iris_resource_check_level_layer(const struct iris_resource *res, + uint32_t level, uint32_t layer); + +bool iris_resource_level_has_hiz(const struct iris_resource *res, + uint32_t level); + +enum isl_aux_usage iris_resource_render_aux_usage(struct iris_context *ice, + struct iris_resource *res, + enum isl_format render_fmt, + bool blend_enabled, + bool draw_aux_disabled); +void iris_resource_prepare_render(struct iris_context *ice, + struct iris_resource *res, uint32_t level, + uint32_t start_layer, uint32_t layer_count, + enum isl_aux_usage aux_usage); +void iris_resource_finish_render(struct iris_context *ice, + struct iris_resource *res, uint32_t level, + uint32_t start_layer, uint32_t layer_count, + enum isl_aux_usage aux_usage); +void iris_resource_prepare_depth(struct iris_context *ice, + struct iris_resource *res, uint32_t level, + uint32_t start_layer, uint32_t layer_count); +void iris_resource_finish_depth(struct iris_context *ice, + struct iris_resource *res, uint32_t level, + uint32_t start_layer, uint32_t layer_count, + bool depth_written); #endif diff --git a/src/gallium/drivers/iris/iris_state.c b/src/gallium/drivers/iris/iris_state.c index aa16f8517d6..9ff40a0160b 100644 --- a/src/gallium/drivers/iris/iris_state.c +++ b/src/gallium/drivers/iris/iris_state.c @@ -3778,6 +3778,8 @@ iris_populate_binding_table(struct iris_context *ice, return; } + // XXX: use different surface states per aux mode + if (stage == MESA_SHADER_COMPUTE) { /* surface for gl_NumWorkGroups */ struct iris_state_ref *grid_data = &ice->state.grid_size; |