* IN THE SOFTWARE.
*/
-#include <errno.h>
-#include "intel_batchbuffer.h"
-#include "intel_fbo.h"
+#include "main/context.h"
+#include "main/teximage.h"
+#include "main/blend.h"
+#include "main/fbobject.h"
+#include "main/renderbuffer.h"
+#include "main/glformats.h"
#include "brw_blorp.h"
+#include "brw_context.h"
+#include "brw_defines.h"
+#include "brw_meta_util.h"
#include "brw_state.h"
+#include "intel_fbo.h"
+#include "common/gen_debug.h"
#define FILE_DEBUG_FLAG DEBUG_BLORP
+static bool
+brw_blorp_lookup_shader(struct blorp_context *blorp,
+ const void *key, uint32_t key_size,
+ uint32_t *kernel_out, void *prog_data_out)
+{
+ struct brw_context *brw = blorp->driver_ctx;
+ return brw_search_cache(&brw->cache, BRW_CACHE_BLORP_PROG,
+ key, key_size, kernel_out, prog_data_out);
+}
+
+static bool
+brw_blorp_upload_shader(struct blorp_context *blorp,
+ const void *key, uint32_t key_size,
+ const void *kernel, uint32_t kernel_size,
+ const struct brw_stage_prog_data *prog_data,
+ uint32_t prog_data_size,
+ uint32_t *kernel_out, void *prog_data_out)
+{
+ struct brw_context *brw = blorp->driver_ctx;
+ brw_upload_cache(&brw->cache, BRW_CACHE_BLORP_PROG, key, key_size,
+ kernel, kernel_size, prog_data, prog_data_size,
+ kernel_out, prog_data_out);
+ return true;
+}
+
void
-brw_blorp_surface_info_init(struct brw_context *brw,
- struct brw_blorp_surface_info *info,
- struct intel_mipmap_tree *mt,
- unsigned int level, unsigned int layer,
- mesa_format format, bool is_render_target)
+brw_blorp_init(struct brw_context *brw)
+{
+ blorp_init(&brw->blorp, brw, &brw->isl_dev);
+
+ brw->blorp.compiler = brw->screen->compiler;
+
+ switch (brw->gen) {
+ case 4:
+ if (brw->is_g4x) {
+ brw->blorp.exec = gen45_blorp_exec;
+ } else {
+ brw->blorp.exec = gen4_blorp_exec;
+ }
+ break;
+ case 5:
+ brw->blorp.exec = gen5_blorp_exec;
+ break;
+ case 6:
+ brw->blorp.mocs.tex = 0;
+ brw->blorp.mocs.rb = 0;
+ brw->blorp.mocs.vb = 0;
+ brw->blorp.exec = gen6_blorp_exec;
+ break;
+ case 7:
+ brw->blorp.mocs.tex = GEN7_MOCS_L3;
+ brw->blorp.mocs.rb = GEN7_MOCS_L3;
+ brw->blorp.mocs.vb = GEN7_MOCS_L3;
+ if (brw->is_haswell) {
+ brw->blorp.exec = gen75_blorp_exec;
+ } else {
+ brw->blorp.exec = gen7_blorp_exec;
+ }
+ break;
+ case 8:
+ brw->blorp.mocs.tex = BDW_MOCS_WB;
+ brw->blorp.mocs.rb = BDW_MOCS_PTE;
+ brw->blorp.mocs.vb = BDW_MOCS_WB;
+ brw->blorp.exec = gen8_blorp_exec;
+ break;
+ case 9:
+ brw->blorp.mocs.tex = SKL_MOCS_WB;
+ brw->blorp.mocs.rb = SKL_MOCS_PTE;
+ brw->blorp.mocs.vb = SKL_MOCS_WB;
+ brw->blorp.exec = gen9_blorp_exec;
+ break;
+ default:
+ unreachable("Invalid gen");
+ }
+
+ brw->blorp.lookup_shader = brw_blorp_lookup_shader;
+ brw->blorp.upload_shader = brw_blorp_upload_shader;
+}
+
+static void
+apply_gen6_stencil_hiz_offset(struct isl_surf *surf,
+ struct intel_mipmap_tree *mt,
+ uint32_t lod,
+ uint32_t *offset)
+{
+ assert(mt->array_layout == ALL_SLICES_AT_EACH_LOD);
+
+ if (mt->format == MESA_FORMAT_S_UINT8) {
+ /* Note: we can't compute the stencil offset using
+ * intel_miptree_get_aligned_offset(), because the miptree
+ * claims that the region is untiled even though it's W tiled.
+ */
+ *offset = mt->level[lod].level_y * mt->pitch +
+ mt->level[lod].level_x * 64;
+ } else {
+ *offset = intel_miptree_get_aligned_offset(mt,
+ mt->level[lod].level_x,
+ mt->level[lod].level_y);
+ }
+
+ surf->logical_level0_px.width = minify(surf->logical_level0_px.width, lod);
+ surf->logical_level0_px.height = minify(surf->logical_level0_px.height, lod);
+ surf->phys_level0_sa.width = minify(surf->phys_level0_sa.width, lod);
+ surf->phys_level0_sa.height = minify(surf->phys_level0_sa.height, lod);
+ surf->levels = 1;
+ surf->array_pitch_el_rows =
+ ALIGN(surf->phys_level0_sa.height, surf->image_alignment_el.height);
+}
+
+static void
+blorp_surf_for_miptree(struct brw_context *brw,
+ struct blorp_surf *surf,
+ struct intel_mipmap_tree *mt,
+ bool is_render_target,
+ uint32_t safe_aux_usage,
+ unsigned *level,
+ unsigned start_layer, unsigned num_layers,
+ struct isl_surf tmp_surfs[2])
{
- /* Layer is a physical layer, so if this is a 2D multisample array texture
- * using INTEL_MSAA_LAYOUT_UMS or INTEL_MSAA_LAYOUT_CMS, then it had better
- * be a multiple of num_samples.
- */
if (mt->msaa_layout == INTEL_MSAA_LAYOUT_UMS ||
mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS) {
- assert(mt->num_samples <= 1 || layer % mt->num_samples == 0);
+ const unsigned num_samples = MAX2(1, mt->num_samples);
+ for (unsigned i = 0; i < num_layers; i++) {
+ for (unsigned s = 0; s < num_samples; s++) {
+ const unsigned phys_layer = (start_layer + i) * num_samples + s;
+ intel_miptree_check_level_layer(mt, *level, phys_layer);
+ }
+ }
+ } else {
+ for (unsigned i = 0; i < num_layers; i++)
+ intel_miptree_check_level_layer(mt, *level, start_layer + i);
+ }
+
+ intel_miptree_get_isl_surf(brw, mt, &tmp_surfs[0]);
+ surf->surf = &tmp_surfs[0];
+ surf->addr = (struct blorp_address) {
+ .buffer = mt->bo,
+ .offset = mt->offset,
+ .read_domains = is_render_target ? I915_GEM_DOMAIN_RENDER :
+ I915_GEM_DOMAIN_SAMPLER,
+ .write_domain = is_render_target ? I915_GEM_DOMAIN_RENDER : 0,
+ };
+
+ if (brw->gen == 6 && mt->format == MESA_FORMAT_S_UINT8 &&
+ mt->array_layout == ALL_SLICES_AT_EACH_LOD) {
+ /* Sandy bridge stencil and HiZ use this ALL_SLICES_AT_EACH_LOD hack in
+ * order to allow for layered rendering. The hack makes each LOD of the
+ * stencil or HiZ buffer a single tightly packed array surface at some
+ * offset into the surface. Since ISL doesn't know how to deal with the
+ * crazy ALL_SLICES_AT_EACH_LOD layout and since we have to do a manual
+ * offset of it anyway, we might as well do the offset here and keep the
+ * hacks inside the i965 driver.
+ *
+ * See also gen6_depth_stencil_state.c
+ */
+ uint32_t offset;
+ apply_gen6_stencil_hiz_offset(&tmp_surfs[0], mt, *level, &offset);
+ surf->addr.offset += offset;
+ *level = 0;
}
- intel_miptree_check_level_layer(mt, level, layer);
+ struct isl_surf *aux_surf = &tmp_surfs[1];
+ intel_miptree_get_aux_isl_surf(brw, mt, aux_surf, &surf->aux_usage);
+
+ if (surf->aux_usage != ISL_AUX_USAGE_NONE) {
+ if (surf->aux_usage == ISL_AUX_USAGE_HIZ) {
+ /* If we're not going to use it as a depth buffer, resolve HiZ */
+ if (!(safe_aux_usage & (1 << ISL_AUX_USAGE_HIZ))) {
+ for (unsigned i = 0; i < num_layers; i++) {
+ intel_miptree_slice_resolve_depth(brw, mt, *level,
+ start_layer + i);
+
+ /* If we're rendering to it then we'll need a HiZ resolve once
+ * we're done before we can use it with HiZ again.
+ */
+ if (is_render_target)
+ intel_miptree_slice_set_needs_hiz_resolve(mt, *level,
+ start_layer + i);
+ }
+ surf->aux_usage = ISL_AUX_USAGE_NONE;
+ }
+ } else if (!(safe_aux_usage & (1 << surf->aux_usage))) {
+ uint32_t flags = 0;
+ if (safe_aux_usage & (1 << ISL_AUX_USAGE_CCS_E))
+ flags |= INTEL_MIPTREE_IGNORE_CCS_E;
+
+ intel_miptree_resolve_color(brw, mt,
+ *level, start_layer, num_layers, flags);
+
+ assert(!intel_miptree_has_color_unresolved(mt, *level, 1,
+ start_layer, num_layers));
+ surf->aux_usage = ISL_AUX_USAGE_NONE;
+ }
+ }
+
+ if (is_render_target)
+ intel_miptree_used_for_rendering(brw, mt, *level,
+ start_layer, num_layers);
+
+ if (surf->aux_usage != ISL_AUX_USAGE_NONE) {
+ /* We only really need a clear color if we also have an auxiliary
+ * surface. Without one, it does nothing.
+ */
+ surf->clear_color = intel_miptree_get_isl_clear_color(brw, mt);
- info->mt = mt;
- info->level = level;
- info->layer = layer;
- info->width = minify(mt->physical_width0, level - mt->first_level);
- info->height = minify(mt->physical_height0, level - mt->first_level);
+ surf->aux_surf = aux_surf;
+ surf->aux_addr = (struct blorp_address) {
+ .read_domains = is_render_target ? I915_GEM_DOMAIN_RENDER :
+ I915_GEM_DOMAIN_SAMPLER,
+ .write_domain = is_render_target ? I915_GEM_DOMAIN_RENDER : 0,
+ };
- intel_miptree_get_image_offset(mt, level, layer,
- &info->x_offset, &info->y_offset);
+ if (mt->mcs_buf) {
+ surf->aux_addr.buffer = mt->mcs_buf->bo;
+ surf->aux_addr.offset = mt->mcs_buf->offset;
+ } else {
+ assert(surf->aux_usage == ISL_AUX_USAGE_HIZ);
- info->num_samples = mt->num_samples;
- info->array_layout = mt->array_layout;
- info->map_stencil_as_y_tiled = false;
- info->msaa_layout = mt->msaa_layout;
- info->swizzle = SWIZZLE_XYZW;
+ surf->aux_addr.buffer = mt->hiz_buf->aux_base.bo;
+ surf->aux_addr.offset = mt->hiz_buf->aux_base.offset;
- if (format == MESA_FORMAT_NONE)
- format = mt->format;
+ struct intel_mipmap_tree *hiz_mt = mt->hiz_buf->mt;
+ if (hiz_mt) {
+ assert(brw->gen == 6 &&
+ hiz_mt->array_layout == ALL_SLICES_AT_EACH_LOD);
+ /* gen6 requires the HiZ buffer to be manually offset to the
+ * right location. We could fixup the surf but it doesn't
+ * matter since most of those fields don't matter.
+ */
+ apply_gen6_stencil_hiz_offset(aux_surf, hiz_mt, *level,
+ &surf->aux_addr.offset);
+ assert(hiz_mt->pitch == aux_surf->row_pitch);
+ }
+ }
+ } else {
+ surf->aux_addr = (struct blorp_address) {
+ .buffer = NULL,
+ };
+ memset(&surf->clear_color, 0, sizeof(surf->clear_color));
+ }
+ assert((surf->aux_usage == ISL_AUX_USAGE_NONE) ==
+ (surf->aux_addr.buffer == NULL));
+
+ /* ISL wants real levels, not offset ones. */
+ *level -= mt->first_level;
+}
+
+static enum isl_format
+brw_blorp_to_isl_format(struct brw_context *brw, mesa_format format,
+ bool is_render_target)
+{
switch (format) {
+ case MESA_FORMAT_NONE:
+ return ISL_FORMAT_UNSUPPORTED;
case MESA_FORMAT_S_UINT8:
- /* The miptree is a W-tiled stencil buffer. Surface states can't be set
- * up for W tiling, so we'll need to use Y tiling and have the WM
- * program swizzle the coordinates.
- */
- info->map_stencil_as_y_tiled = true;
- info->brw_surfaceformat = brw->gen >= 8 ? BRW_SURFACEFORMAT_R8_UINT :
- BRW_SURFACEFORMAT_R8_UNORM;
- break;
+ return ISL_FORMAT_R8_UINT;
case MESA_FORMAT_Z24_UNORM_X8_UINT:
- /* It would make sense to use BRW_SURFACEFORMAT_R24_UNORM_X8_TYPELESS
- * here, but unfortunately it isn't supported as a render target, which
- * would prevent us from blitting to 24-bit depth.
- *
- * The miptree consists of 32 bits per pixel, arranged as 24-bit depth
- * values interleaved with 8 "don't care" bits. Since depth values don't
- * require any blending, it doesn't matter how we interpret the bit
- * pattern as long as we copy the right amount of data, so just map it
- * as 8-bit BGRA.
- */
- info->brw_surfaceformat = BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
- break;
+ case MESA_FORMAT_Z24_UNORM_S8_UINT:
+ return ISL_FORMAT_R24_UNORM_X8_TYPELESS;
case MESA_FORMAT_Z_FLOAT32:
- info->brw_surfaceformat = BRW_SURFACEFORMAT_R32_FLOAT;
- break;
+ case MESA_FORMAT_Z32_FLOAT_S8X24_UINT:
+ return ISL_FORMAT_R32_FLOAT;
case MESA_FORMAT_Z_UNORM16:
- info->brw_surfaceformat = BRW_SURFACEFORMAT_R16_UNORM;
- break;
+ return ISL_FORMAT_R16_UNORM;
default: {
if (is_render_target) {
assert(brw->format_supported_as_render_target[format]);
- info->brw_surfaceformat = brw->render_target_format[format];
+ return brw->render_target_format[format];
} else {
- info->brw_surfaceformat = brw_format_for_mesa_format(format);
+ return brw_isl_format_for_mesa_format(format);
}
break;
}
}
}
+/**
+ * Convert an swizzle enumeration (i.e. SWIZZLE_X) to one of the Gen7.5+
+ * "Shader Channel Select" enumerations (i.e. HSW_SCS_RED). The mappings are
+ *
+ * SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_W, SWIZZLE_ZERO, SWIZZLE_ONE
+ * 0 1 2 3 4 5
+ * 4 5 6 7 0 1
+ * SCS_RED, SCS_GREEN, SCS_BLUE, SCS_ALPHA, SCS_ZERO, SCS_ONE
+ *
+ * which is simply adding 4 then modding by 8 (or anding with 7).
+ *
+ * We then may need to apply workarounds for textureGather hardware bugs.
+ */
+static enum isl_channel_select
+swizzle_to_scs(GLenum swizzle)
+{
+ return (enum isl_channel_select)((swizzle + 4) & 7);
+}
+
+static unsigned
+physical_to_logical_layer(struct intel_mipmap_tree *mt,
+ unsigned physical_layer)
+{
+ if (mt->num_samples > 1 &&
+ (mt->msaa_layout == INTEL_MSAA_LAYOUT_UMS ||
+ mt->msaa_layout == INTEL_MSAA_LAYOUT_CMS)) {
+ assert(physical_layer % mt->num_samples == 0);
+ return physical_layer / mt->num_samples;
+ } else {
+ return physical_layer;
+ }
+}
/**
- * Split x_offset and y_offset into a base offset (in bytes) and a remaining
- * x/y offset (in pixels). Note: we can't do this by calling
- * intel_renderbuffer_tile_offsets(), because the offsets may have been
- * adjusted to account for Y vs. W tiling differences. So we compute it
- * directly from the adjusted offsets.
+ * Note: if the src (or dst) is a 2D multisample array texture on Gen7+ using
+ * INTEL_MSAA_LAYOUT_UMS or INTEL_MSAA_LAYOUT_CMS, src_layer (dst_layer) is
+ * the physical layer holding sample 0. So, for example, if
+ * src_mt->num_samples == 4, then logical layer n corresponds to src_layer ==
+ * 4*n.
*/
-uint32_t
-brw_blorp_compute_tile_offsets(const struct brw_blorp_surface_info *info,
- uint32_t *tile_x, uint32_t *tile_y)
+void
+brw_blorp_blit_miptrees(struct brw_context *brw,
+ struct intel_mipmap_tree *src_mt,
+ unsigned src_level, unsigned src_layer,
+ mesa_format src_format, int src_swizzle,
+ struct intel_mipmap_tree *dst_mt,
+ unsigned dst_level, unsigned dst_layer,
+ mesa_format dst_format,
+ float src_x0, float src_y0,
+ float src_x1, float src_y1,
+ float dst_x0, float dst_y0,
+ float dst_x1, float dst_y1,
+ GLenum filter, bool mirror_x, bool mirror_y,
+ bool decode_srgb, bool encode_srgb)
{
- uint32_t mask_x, mask_y;
+ /* Blorp operates in logical layers */
+ src_layer = physical_to_logical_layer(src_mt, src_layer);
+ dst_layer = physical_to_logical_layer(dst_mt, dst_layer);
- intel_get_tile_masks(info->mt->tiling, info->mt->tr_mode, info->mt->cpp,
- info->map_stencil_as_y_tiled,
- &mask_x, &mask_y);
+ DBG("%s from %dx %s mt %p %d %d (%f,%f) (%f,%f)"
+ "to %dx %s mt %p %d %d (%f,%f) (%f,%f) (flip %d,%d)\n",
+ __func__,
+ src_mt->num_samples, _mesa_get_format_name(src_mt->format), src_mt,
+ src_level, src_layer, src_x0, src_y0, src_x1, src_y1,
+ dst_mt->num_samples, _mesa_get_format_name(dst_mt->format), dst_mt,
+ dst_level, dst_layer, dst_x0, dst_y0, dst_x1, dst_y1,
+ mirror_x, mirror_y);
- *tile_x = info->x_offset & mask_x;
- *tile_y = info->y_offset & mask_y;
+ if (!decode_srgb && _mesa_get_format_color_encoding(src_format) == GL_SRGB)
+ src_format = _mesa_get_srgb_format_linear(src_format);
- return intel_miptree_get_aligned_offset(info->mt, info->x_offset & ~mask_x,
- info->y_offset & ~mask_y,
- info->map_stencil_as_y_tiled);
-}
+ if (!encode_srgb && _mesa_get_format_color_encoding(dst_format) == GL_SRGB)
+ dst_format = _mesa_get_srgb_format_linear(dst_format);
+
+ /* When doing a multisample resolve of a GL_LUMINANCE32F or GL_INTENSITY32F
+ * texture, the above code configures the source format for L32_FLOAT or
+ * I32_FLOAT, and the destination format for R32_FLOAT. On Sandy Bridge,
+ * the SAMPLE message appears to handle multisampled L32_FLOAT and
+ * I32_FLOAT textures incorrectly, resulting in blocky artifacts. So work
+ * around the problem by using a source format of R32_FLOAT. This
+ * shouldn't affect rendering correctness, since the destination format is
+ * R32_FLOAT, so only the contents of the red channel matters.
+ */
+ if (brw->gen == 6 &&
+ src_mt->num_samples > 1 && dst_mt->num_samples <= 1 &&
+ src_mt->format == dst_mt->format &&
+ (dst_format == MESA_FORMAT_L_FLOAT32 ||
+ dst_format == MESA_FORMAT_I_FLOAT32)) {
+ src_format = dst_format = MESA_FORMAT_R_FLOAT32;
+ }
+
+ uint32_t src_usage_flags = (1 << ISL_AUX_USAGE_MCS);
+ if (src_format == src_mt->format)
+ src_usage_flags |= (1 << ISL_AUX_USAGE_CCS_E);
+
+ uint32_t dst_usage_flags = (1 << ISL_AUX_USAGE_MCS);
+ if (dst_format == dst_mt->format) {
+ dst_usage_flags |= (1 << ISL_AUX_USAGE_CCS_E) |
+ (1 << ISL_AUX_USAGE_CCS_D);
+ }
+ struct isl_surf tmp_surfs[4];
+ struct blorp_surf src_surf, dst_surf;
+ blorp_surf_for_miptree(brw, &src_surf, src_mt, false, src_usage_flags,
+ &src_level, src_layer, 1, &tmp_surfs[0]);
+ blorp_surf_for_miptree(brw, &dst_surf, dst_mt, true, dst_usage_flags,
+ &dst_level, dst_layer, 1, &tmp_surfs[2]);
+
+ struct isl_swizzle src_isl_swizzle = {
+ .r = swizzle_to_scs(GET_SWZ(src_swizzle, 0)),
+ .g = swizzle_to_scs(GET_SWZ(src_swizzle, 1)),
+ .b = swizzle_to_scs(GET_SWZ(src_swizzle, 2)),
+ .a = swizzle_to_scs(GET_SWZ(src_swizzle, 3)),
+ };
+
+ struct blorp_batch batch;
+ blorp_batch_init(&brw->blorp, &batch, brw, 0);
+ blorp_blit(&batch, &src_surf, src_level, src_layer,
+ brw_blorp_to_isl_format(brw, src_format, false), src_isl_swizzle,
+ &dst_surf, dst_level, dst_layer,
+ brw_blorp_to_isl_format(brw, dst_format, true),
+ ISL_SWIZZLE_IDENTITY,
+ src_x0, src_y0, src_x1, src_y1,
+ dst_x0, dst_y0, dst_x1, dst_y1,
+ filter, mirror_x, mirror_y);
+ blorp_batch_finish(&batch);
+}
void
-brw_blorp_prog_data_init(struct brw_blorp_prog_data *prog_data)
+brw_blorp_copy_miptrees(struct brw_context *brw,
+ struct intel_mipmap_tree *src_mt,
+ unsigned src_level, unsigned src_layer,
+ struct intel_mipmap_tree *dst_mt,
+ unsigned dst_level, unsigned dst_layer,
+ unsigned src_x, unsigned src_y,
+ unsigned dst_x, unsigned dst_y,
+ unsigned src_width, unsigned src_height)
{
- prog_data->dispatch_8 = false;
- prog_data->dispatch_16 = true;
- prog_data->first_curbe_grf_0 = 0;
- prog_data->first_curbe_grf_2 = 0;
- prog_data->ksp_offset_2 = 0;
- prog_data->persample_msaa_dispatch = false;
-
- prog_data->nr_params = BRW_BLORP_NUM_PUSH_CONSTANT_DWORDS;
- for (unsigned i = 0; i < BRW_BLORP_NUM_PUSH_CONSTANT_DWORDS; i++)
- prog_data->param[i] = i;
+ DBG("%s from %dx %s mt %p %d %d (%d,%d) %dx%d"
+ "to %dx %s mt %p %d %d (%d,%d)\n",
+ __func__,
+ src_mt->num_samples, _mesa_get_format_name(src_mt->format), src_mt,
+ src_level, src_layer, src_x, src_y, src_width, src_height,
+ dst_mt->num_samples, _mesa_get_format_name(dst_mt->format), dst_mt,
+ dst_level, dst_layer, dst_x, dst_y);
+
+ struct isl_surf tmp_surfs[4];
+ struct blorp_surf src_surf, dst_surf;
+ blorp_surf_for_miptree(brw, &src_surf, src_mt, false,
+ (1 << ISL_AUX_USAGE_MCS) |
+ (1 << ISL_AUX_USAGE_CCS_E),
+ &src_level, src_layer, 1, &tmp_surfs[0]);
+ blorp_surf_for_miptree(brw, &dst_surf, dst_mt, true,
+ (1 << ISL_AUX_USAGE_MCS) |
+ (1 << ISL_AUX_USAGE_CCS_E),
+ &dst_level, dst_layer, 1, &tmp_surfs[2]);
+
+ struct blorp_batch batch;
+ blorp_batch_init(&brw->blorp, &batch, brw, 0);
+ blorp_copy(&batch, &src_surf, src_level, src_layer,
+ &dst_surf, dst_level, dst_layer,
+ src_x, src_y, dst_x, dst_y, src_width, src_height);
+ blorp_batch_finish(&batch);
}
+static struct intel_mipmap_tree *
+find_miptree(GLbitfield buffer_bit, struct intel_renderbuffer *irb)
+{
+ struct intel_mipmap_tree *mt = irb->mt;
+ if (buffer_bit == GL_STENCIL_BUFFER_BIT && mt->stencil_mt)
+ mt = mt->stencil_mt;
+ return mt;
+}
-void
-brw_blorp_params_init(struct brw_blorp_params *params)
+static int
+blorp_get_texture_swizzle(const struct intel_renderbuffer *irb)
{
- memset(params, 0, sizeof(*params));
- params->hiz_op = GEN6_HIZ_OP_NONE;
- params->fast_clear_op = 0;
- params->num_varyings = 0;
- params->num_draw_buffers = 1;
- params->num_layers = 1;
+ return irb->Base.Base._BaseFormat == GL_RGB ?
+ MAKE_SWIZZLE4(SWIZZLE_X, SWIZZLE_Y, SWIZZLE_Z, SWIZZLE_ONE) :
+ SWIZZLE_XYZW;
}
-/**
- * Perform a HiZ or depth resolve operation.
- *
- * For an overview of HiZ ops, see the following sections of the Sandy Bridge
- * PRM, Volume 1, Part 2:
- * - 7.5.3.1 Depth Buffer Clear
- * - 7.5.3.2 Depth Buffer Resolve
- * - 7.5.3.3 Hierarchical Depth Buffer Resolve
- */
-void
-intel_hiz_exec(struct brw_context *brw, struct intel_mipmap_tree *mt,
- unsigned int level, unsigned int layer, enum gen6_hiz_op op)
+static void
+do_blorp_blit(struct brw_context *brw, GLbitfield buffer_bit,
+ struct intel_renderbuffer *src_irb, mesa_format src_format,
+ struct intel_renderbuffer *dst_irb, mesa_format dst_format,
+ GLfloat srcX0, GLfloat srcY0, GLfloat srcX1, GLfloat srcY1,
+ GLfloat dstX0, GLfloat dstY0, GLfloat dstX1, GLfloat dstY1,
+ GLenum filter, bool mirror_x, bool mirror_y)
{
- const char *opname = NULL;
+ const struct gl_context *ctx = &brw->ctx;
- switch (op) {
- case GEN6_HIZ_OP_DEPTH_RESOLVE:
- opname = "depth resolve";
- break;
- case GEN6_HIZ_OP_HIZ_RESOLVE:
- opname = "hiz ambiguate";
- break;
- case GEN6_HIZ_OP_DEPTH_CLEAR:
- opname = "depth clear";
- break;
- case GEN6_HIZ_OP_NONE:
- opname = "noop?";
- break;
- }
+ /* Find source/dst miptrees */
+ struct intel_mipmap_tree *src_mt = find_miptree(buffer_bit, src_irb);
+ struct intel_mipmap_tree *dst_mt = find_miptree(buffer_bit, dst_irb);
- DBG("%s %s to mt %p level %d layer %d\n",
- __func__, opname, mt, level, layer);
+ const bool do_srgb = ctx->Color.sRGBEnabled;
- if (brw->gen >= 8) {
- gen8_hiz_exec(brw, mt, level, layer, op);
- } else {
- gen6_blorp_hiz_exec(brw, mt, level, layer, op);
- }
+ /* Do the blit */
+ brw_blorp_blit_miptrees(brw,
+ src_mt, src_irb->mt_level, src_irb->mt_layer,
+ src_format, blorp_get_texture_swizzle(src_irb),
+ dst_mt, dst_irb->mt_level, dst_irb->mt_layer,
+ dst_format,
+ srcX0, srcY0, srcX1, srcY1,
+ dstX0, dstY0, dstX1, dstY1,
+ filter, mirror_x, mirror_y,
+ do_srgb, do_srgb);
+
+ dst_irb->need_downsample = true;
}
-void
-brw_blorp_exec(struct brw_context *brw, const struct brw_blorp_params *params)
+static bool
+try_blorp_blit(struct brw_context *brw,
+ const struct gl_framebuffer *read_fb,
+ const struct gl_framebuffer *draw_fb,
+ GLfloat srcX0, GLfloat srcY0, GLfloat srcX1, GLfloat srcY1,
+ GLfloat dstX0, GLfloat dstY0, GLfloat dstX1, GLfloat dstY1,
+ GLenum filter, GLbitfield buffer_bit)
{
struct gl_context *ctx = &brw->ctx;
- const uint32_t estimated_max_batch_usage = brw->gen >= 8 ? 1800 : 1500;
- bool check_aperture_failed_once = false;
-
- /* Flush the sampler and render caches. We definitely need to flush the
- * sampler cache so that we get updated contents from the render cache for
- * the glBlitFramebuffer() source. Also, we are sometimes warned in the
- * docs to flush the cache between reinterpretations of the same surface
- * data with different formats, which blorp does for stencil and depth
- * data.
- */
- brw_emit_mi_flush(brw);
- brw_select_pipeline(brw, BRW_RENDER_PIPELINE);
+ /* Sync up the state of window system buffers. We need to do this before
+ * we go looking for the buffers.
+ */
+ intel_prepare_render(brw);
-retry:
- intel_batchbuffer_require_space(brw, estimated_max_batch_usage, RENDER_RING);
- intel_batchbuffer_save_state(brw);
- drm_intel_bo *saved_bo = brw->batch.bo;
- uint32_t saved_used = USED_BATCH(brw->batch);
- uint32_t saved_state_batch_offset = brw->batch.state_batch_offset;
+ bool mirror_x, mirror_y;
+ if (brw_meta_mirror_clip_and_scissor(ctx, read_fb, draw_fb,
+ &srcX0, &srcY0, &srcX1, &srcY1,
+ &dstX0, &dstY0, &dstX1, &dstY1,
+ &mirror_x, &mirror_y))
+ return true;
- switch (brw->gen) {
- case 6:
- gen6_blorp_exec(brw, params);
+ /* Find buffers */
+ struct intel_renderbuffer *src_irb;
+ struct intel_renderbuffer *dst_irb;
+ struct intel_mipmap_tree *src_mt;
+ struct intel_mipmap_tree *dst_mt;
+ switch (buffer_bit) {
+ case GL_COLOR_BUFFER_BIT:
+ src_irb = intel_renderbuffer(read_fb->_ColorReadBuffer);
+ for (unsigned i = 0; i < draw_fb->_NumColorDrawBuffers; ++i) {
+ dst_irb = intel_renderbuffer(draw_fb->_ColorDrawBuffers[i]);
+ if (dst_irb)
+ do_blorp_blit(brw, buffer_bit,
+ src_irb, src_irb->Base.Base.Format,
+ dst_irb, dst_irb->Base.Base.Format,
+ srcX0, srcY0, srcX1, srcY1,
+ dstX0, dstY0, dstX1, dstY1,
+ filter, mirror_x, mirror_y);
+ }
break;
- case 7:
- gen7_blorp_exec(brw, params);
+ case GL_DEPTH_BUFFER_BIT:
+ src_irb =
+ intel_renderbuffer(read_fb->Attachment[BUFFER_DEPTH].Renderbuffer);
+ dst_irb =
+ intel_renderbuffer(draw_fb->Attachment[BUFFER_DEPTH].Renderbuffer);
+ src_mt = find_miptree(buffer_bit, src_irb);
+ dst_mt = find_miptree(buffer_bit, dst_irb);
+
+ /* We can't handle format conversions between Z24 and other formats
+ * since we have to lie about the surface format. See the comments in
+ * brw_blorp_surface_info::set().
+ */
+ if ((src_mt->format == MESA_FORMAT_Z24_UNORM_X8_UINT) !=
+ (dst_mt->format == MESA_FORMAT_Z24_UNORM_X8_UINT))
+ return false;
+
+ /* We also can't handle any combined depth-stencil formats because we
+ * have to reinterpret as a color format.
+ */
+ if (_mesa_get_format_base_format(src_mt->format) == GL_DEPTH_STENCIL ||
+ _mesa_get_format_base_format(dst_mt->format) == GL_DEPTH_STENCIL)
+ return false;
+
+ do_blorp_blit(brw, buffer_bit, src_irb, MESA_FORMAT_NONE,
+ dst_irb, MESA_FORMAT_NONE, srcX0, srcY0,
+ srcX1, srcY1, dstX0, dstY0, dstX1, dstY1,
+ filter, mirror_x, mirror_y);
break;
- case 8:
- case 9:
- gen8_blorp_exec(brw, params);
+ case GL_STENCIL_BUFFER_BIT:
+ /* Blorp doesn't support combined depth stencil which is all we have
+ * prior to gen6.
+ */
+ if (brw->gen < 6)
+ return false;
+
+ src_irb =
+ intel_renderbuffer(read_fb->Attachment[BUFFER_STENCIL].Renderbuffer);
+ dst_irb =
+ intel_renderbuffer(draw_fb->Attachment[BUFFER_STENCIL].Renderbuffer);
+ do_blorp_blit(brw, buffer_bit, src_irb, MESA_FORMAT_NONE,
+ dst_irb, MESA_FORMAT_NONE, srcX0, srcY0,
+ srcX1, srcY1, dstX0, dstY0, dstX1, dstY1,
+ filter, mirror_x, mirror_y);
break;
default:
- /* BLORP is not supported before Gen6. */
unreachable("not reached");
}
- /* Make sure we didn't wrap the batch unintentionally, and make sure we
- * reserved enough space that a wrap will never happen.
+ return true;
+}
+
+bool
+brw_blorp_copytexsubimage(struct brw_context *brw,
+ struct gl_renderbuffer *src_rb,
+ struct gl_texture_image *dst_image,
+ int slice,
+ int srcX0, int srcY0,
+ int dstX0, int dstY0,
+ int width, int height)
+{
+ struct gl_context *ctx = &brw->ctx;
+ struct intel_renderbuffer *src_irb = intel_renderbuffer(src_rb);
+ struct intel_texture_image *intel_image = intel_texture_image(dst_image);
+
+ /* No pixel transfer operations (zoom, bias, mapping), just a blit */
+ if (brw->ctx._ImageTransferState)
+ return false;
+
+ /* Sync up the state of window system buffers. We need to do this before
+ * we go looking at the src renderbuffer's miptree.
+ */
+ intel_prepare_render(brw);
+
+ struct intel_mipmap_tree *src_mt = src_irb->mt;
+ struct intel_mipmap_tree *dst_mt = intel_image->mt;
+
+ /* There is support for only up to eight samples. */
+ if (src_mt->num_samples > 8 || dst_mt->num_samples > 8)
+ return false;
+
+ if (_mesa_get_format_base_format(src_rb->Format) !=
+ _mesa_get_format_base_format(dst_image->TexFormat)) {
+ return false;
+ }
+
+ /* We can't handle format conversions between Z24 and other formats since
+ * we have to lie about the surface format. See the comments in
+ * brw_blorp_surface_info::set().
*/
- assert(brw->batch.bo == saved_bo);
- assert((USED_BATCH(brw->batch) - saved_used) * 4 +
- (saved_state_batch_offset - brw->batch.state_batch_offset) <
- estimated_max_batch_usage);
- /* Shut up compiler warnings on release build */
- (void)saved_bo;
- (void)saved_used;
- (void)saved_state_batch_offset;
-
- /* Check if the blorp op we just did would make our batch likely to fail to
- * map all the BOs into the GPU at batch exec time later. If so, flush the
- * batch and try again with nothing else in the batch.
+ if ((src_mt->format == MESA_FORMAT_Z24_UNORM_X8_UINT) !=
+ (dst_mt->format == MESA_FORMAT_Z24_UNORM_X8_UINT)) {
+ return false;
+ }
+
+ /* We also can't handle any combined depth-stencil formats because we
+ * have to reinterpret as a color format.
*/
- if (dri_bufmgr_check_aperture_space(&brw->batch.bo, 1)) {
- if (!check_aperture_failed_once) {
- check_aperture_failed_once = true;
- intel_batchbuffer_reset_to_saved(brw);
- intel_batchbuffer_flush(brw);
- goto retry;
- } else {
- int ret = intel_batchbuffer_flush(brw);
- WARN_ONCE(ret == -ENOSPC,
- "i965: blorp emit exceeded available aperture space\n");
+ if (_mesa_get_format_base_format(src_mt->format) == GL_DEPTH_STENCIL ||
+ _mesa_get_format_base_format(dst_mt->format) == GL_DEPTH_STENCIL)
+ return false;
+
+ if (!brw->format_supported_as_render_target[dst_image->TexFormat])
+ return false;
+
+ /* Source clipping shouldn't be necessary, since copytexsubimage (in
+ * src/mesa/main/teximage.c) calls _mesa_clip_copytexsubimage() which
+ * takes care of it.
+ *
+ * Destination clipping shouldn't be necessary since the restrictions on
+ * glCopyTexSubImage prevent the user from specifying a destination rectangle
+ * that falls outside the bounds of the destination texture.
+ * See error_check_subtexture_dimensions().
+ */
+
+ int srcY1 = srcY0 + height;
+ int srcX1 = srcX0 + width;
+ int dstX1 = dstX0 + width;
+ int dstY1 = dstY0 + height;
+
+ /* Account for the fact that in the system framebuffer, the origin is at
+ * the lower left.
+ */
+ bool mirror_y = false;
+ if (_mesa_is_winsys_fbo(ctx->ReadBuffer)) {
+ GLint tmp = src_rb->Height - srcY0;
+ srcY0 = src_rb->Height - srcY1;
+ srcY1 = tmp;
+ mirror_y = true;
+ }
+
+ /* Account for face selection and texture view MinLayer */
+ int dst_slice = slice + dst_image->TexObject->MinLayer + dst_image->Face;
+ int dst_level = dst_image->Level + dst_image->TexObject->MinLevel;
+
+ brw_blorp_blit_miptrees(brw,
+ src_mt, src_irb->mt_level, src_irb->mt_layer,
+ src_rb->Format, blorp_get_texture_swizzle(src_irb),
+ dst_mt, dst_level, dst_slice,
+ dst_image->TexFormat,
+ srcX0, srcY0, srcX1, srcY1,
+ dstX0, dstY0, dstX1, dstY1,
+ GL_NEAREST, false, mirror_y,
+ false, false);
+
+ /* If we're copying to a packed depth stencil texture and the source
+ * framebuffer has separate stencil, we need to also copy the stencil data
+ * over.
+ */
+ src_rb = ctx->ReadBuffer->Attachment[BUFFER_STENCIL].Renderbuffer;
+ if (_mesa_get_format_bits(dst_image->TexFormat, GL_STENCIL_BITS) > 0 &&
+ src_rb != NULL) {
+ src_irb = intel_renderbuffer(src_rb);
+ src_mt = src_irb->mt;
+
+ if (src_mt->stencil_mt)
+ src_mt = src_mt->stencil_mt;
+ if (dst_mt->stencil_mt)
+ dst_mt = dst_mt->stencil_mt;
+
+ if (src_mt != dst_mt) {
+ brw_blorp_blit_miptrees(brw,
+ src_mt, src_irb->mt_level, src_irb->mt_layer,
+ src_mt->format,
+ blorp_get_texture_swizzle(src_irb),
+ dst_mt, dst_level, dst_slice,
+ dst_mt->format,
+ srcX0, srcY0, srcX1, srcY1,
+ dstX0, dstY0, dstX1, dstY1,
+ GL_NEAREST, false, mirror_y,
+ false, false);
+ }
+ }
+
+ return true;
+}
+
+
+GLbitfield
+brw_blorp_framebuffer(struct brw_context *brw,
+ struct gl_framebuffer *readFb,
+ struct gl_framebuffer *drawFb,
+ GLint srcX0, GLint srcY0, GLint srcX1, GLint srcY1,
+ GLint dstX0, GLint dstY0, GLint dstX1, GLint dstY1,
+ GLbitfield mask, GLenum filter)
+{
+ static GLbitfield buffer_bits[] = {
+ GL_COLOR_BUFFER_BIT,
+ GL_DEPTH_BUFFER_BIT,
+ GL_STENCIL_BUFFER_BIT,
+ };
+
+ for (unsigned int i = 0; i < ARRAY_SIZE(buffer_bits); ++i) {
+ if ((mask & buffer_bits[i]) &&
+ try_blorp_blit(brw, readFb, drawFb,
+ srcX0, srcY0, srcX1, srcY1,
+ dstX0, dstY0, dstX1, dstY1,
+ filter, buffer_bits[i])) {
+ mask &= ~buffer_bits[i];
}
}
- if (unlikely(brw->always_flush_batch))
- intel_batchbuffer_flush(brw);
+ return mask;
+}
- /* We've smashed all state compared to what the normal 3D pipeline
- * rendering tracks for GL.
+static bool
+set_write_disables(const struct intel_renderbuffer *irb,
+ const GLubyte *color_mask, bool *color_write_disable)
+{
+ /* Format information in the renderbuffer represents the requirements
+ * given by the client. There are cases where the backing miptree uses,
+ * for example, RGBA to represent RGBX. Since the client is only expecting
+ * RGB we can treat alpha as not used and write whatever we like into it.
*/
- brw->ctx.NewDriverState |= BRW_NEW_BLORP;
- brw->no_depth_or_stencil = false;
- brw->ib.type = -1;
+ const GLenum base_format = irb->Base.Base._BaseFormat;
+ const int components = _mesa_base_format_component_count(base_format);
+ bool disables = false;
+
+ assert(components > 0);
+
+ for (int i = 0; i < components; i++) {
+ color_write_disable[i] = !color_mask[i];
+ disables = disables || !color_mask[i];
+ }
+
+ return disables;
+}
+
+static unsigned
+irb_logical_mt_layer(struct intel_renderbuffer *irb)
+{
+ return physical_to_logical_layer(irb->mt, irb->mt_layer);
+}
+
+static bool
+do_single_blorp_clear(struct brw_context *brw, struct gl_framebuffer *fb,
+ struct gl_renderbuffer *rb, unsigned buf,
+ bool partial_clear, bool encode_srgb)
+{
+ struct gl_context *ctx = &brw->ctx;
+ struct intel_renderbuffer *irb = intel_renderbuffer(rb);
+ mesa_format format = irb->mt->format;
+ uint32_t x0, x1, y0, y1;
+
+ if (!encode_srgb && _mesa_get_format_color_encoding(format) == GL_SRGB)
+ format = _mesa_get_srgb_format_linear(format);
+
+ x0 = fb->_Xmin;
+ x1 = fb->_Xmax;
+ if (rb->Name != 0) {
+ y0 = fb->_Ymin;
+ y1 = fb->_Ymax;
+ } else {
+ y0 = rb->Height - fb->_Ymax;
+ y1 = rb->Height - fb->_Ymin;
+ }
- /* Flush the sampler cache so any texturing from the destination is
- * coherent.
+ /* If the clear region is empty, just return. */
+ if (x0 == x1 || y0 == y1)
+ return true;
+
+ bool can_fast_clear = !partial_clear;
+
+ bool color_write_disable[4] = { false, false, false, false };
+ if (set_write_disables(irb, ctx->Color.ColorMask[buf], color_write_disable))
+ can_fast_clear = false;
+
+ if (irb->mt->aux_disable & INTEL_AUX_DISABLE_CCS ||
+ !brw_is_color_fast_clear_compatible(brw, irb->mt, &ctx->Color.ClearColor))
+ can_fast_clear = false;
+
+ const unsigned logical_layer = irb_logical_mt_layer(irb);
+ const enum intel_fast_clear_state fast_clear_state =
+ intel_miptree_get_fast_clear_state(irb->mt, irb->mt_level,
+ logical_layer);
+
+ /* Surface state can only record one fast clear color value. Therefore
+ * unless different levels/layers agree on the color it can be used to
+ * represent only single level/layer. Here it will be reserved for the
+ * first slice (level 0, layer 0).
*/
- brw_emit_mi_flush(brw);
+ if (irb->layer_count > 1 || irb->mt_level || irb->mt_layer)
+ can_fast_clear = false;
+
+ if (can_fast_clear) {
+ union gl_color_union override_color =
+ brw_meta_convert_fast_clear_color(brw, irb->mt,
+ &ctx->Color.ClearColor);
+
+ /* Record the clear color in the miptree so that it will be
+ * programmed in SURFACE_STATE by later rendering and resolve
+ * operations.
+ */
+ const bool color_updated = brw_meta_set_fast_clear_color(
+ brw, &irb->mt->gen9_fast_clear_color,
+ &override_color);
+
+ /* If the buffer is already in INTEL_FAST_CLEAR_STATE_CLEAR, the clear
+ * is redundant and can be skipped.
+ */
+ if (!color_updated && fast_clear_state == INTEL_FAST_CLEAR_STATE_CLEAR)
+ return true;
+
+ /* If the MCS buffer hasn't been allocated yet, we need to allocate
+ * it now.
+ */
+ if (!irb->mt->mcs_buf) {
+ assert(!intel_miptree_is_lossless_compressed(brw, irb->mt));
+ if (!intel_miptree_alloc_non_msrt_mcs(brw, irb->mt, false)) {
+ /* MCS allocation failed--probably this will only happen in
+ * out-of-memory conditions. But in any case, try to recover
+ * by falling back to a non-blorp clear technique.
+ */
+ return false;
+ }
+ }
+ }
+
+ const unsigned num_layers = fb->MaxNumLayers ? irb->layer_count : 1;
+
+ /* We can't setup the blorp_surf until we've allocated the MCS above */
+ struct isl_surf isl_tmp[2];
+ struct blorp_surf surf;
+ unsigned level = irb->mt_level;
+ blorp_surf_for_miptree(brw, &surf, irb->mt, true,
+ (1 << ISL_AUX_USAGE_MCS) |
+ (1 << ISL_AUX_USAGE_CCS_E) |
+ (1 << ISL_AUX_USAGE_CCS_D),
+ &level, logical_layer, num_layers, isl_tmp);
+
+ if (can_fast_clear) {
+ DBG("%s (fast) to mt %p level %d layers %d+%d\n", __FUNCTION__,
+ irb->mt, irb->mt_level, irb->mt_layer, num_layers);
+
+ /* Ivybrigde PRM Vol 2, Part 1, "11.7 MCS Buffer for Render Target(s)":
+ *
+ * "Any transition from any value in {Clear, Render, Resolve} to a
+ * different value in {Clear, Render, Resolve} requires end of pipe
+ * synchronization."
+ *
+ * In other words, fast clear ops are not properly synchronized with
+ * other drawing. We need to use a PIPE_CONTROL to ensure that the
+ * contents of the previous draw hit the render target before we resolve
+ * and again afterwards to ensure that the resolve is complete before we
+ * do any more regular drawing.
+ */
+ brw_emit_pipe_control_flush(brw,
+ PIPE_CONTROL_RENDER_TARGET_FLUSH |
+ PIPE_CONTROL_CS_STALL);
+
+ struct blorp_batch batch;
+ blorp_batch_init(&brw->blorp, &batch, brw, 0);
+ blorp_fast_clear(&batch, &surf,
+ brw->render_target_format[format],
+ level, logical_layer, num_layers,
+ x0, y0, x1, y1);
+ blorp_batch_finish(&batch);
+
+ brw_emit_pipe_control_flush(brw,
+ PIPE_CONTROL_RENDER_TARGET_FLUSH |
+ PIPE_CONTROL_CS_STALL);
+
+ /* Now that the fast clear has occurred, put the buffer in
+ * INTEL_FAST_CLEAR_STATE_CLEAR so that we won't waste time doing
+ * redundant clears.
+ */
+ intel_miptree_set_fast_clear_state(brw, irb->mt, irb->mt_level,
+ logical_layer, num_layers,
+ INTEL_FAST_CLEAR_STATE_CLEAR);
+ } else {
+ DBG("%s (slow) to mt %p level %d layer %d+%d\n", __FUNCTION__,
+ irb->mt, irb->mt_level, irb->mt_layer, num_layers);
+
+ union isl_color_value clear_color;
+ memcpy(clear_color.f32, ctx->Color.ClearColor.f, sizeof(float) * 4);
+
+ struct blorp_batch batch;
+ blorp_batch_init(&brw->blorp, &batch, brw, 0);
+ blorp_clear(&batch, &surf,
+ brw->render_target_format[format],
+ ISL_SWIZZLE_IDENTITY,
+ level, irb_logical_mt_layer(irb), num_layers,
+ x0, y0, x1, y1,
+ clear_color, color_write_disable);
+ blorp_batch_finish(&batch);
+ }
+
+ return true;
+}
+
+bool
+brw_blorp_clear_color(struct brw_context *brw, struct gl_framebuffer *fb,
+ GLbitfield mask, bool partial_clear, bool encode_srgb)
+{
+ for (unsigned buf = 0; buf < fb->_NumColorDrawBuffers; buf++) {
+ struct gl_renderbuffer *rb = fb->_ColorDrawBuffers[buf];
+ struct intel_renderbuffer *irb = intel_renderbuffer(rb);
+
+ /* Only clear the buffers present in the provided mask */
+ if (((1 << fb->_ColorDrawBufferIndexes[buf]) & mask) == 0)
+ continue;
+
+ /* If this is an ES2 context or GL_ARB_ES2_compatibility is supported,
+ * the framebuffer can be complete with some attachments missing. In
+ * this case the _ColorDrawBuffers pointer will be NULL.
+ */
+ if (rb == NULL)
+ continue;
+
+ if (!do_single_blorp_clear(brw, fb, rb, buf, partial_clear,
+ encode_srgb)) {
+ return false;
+ }
+
+ irb->need_downsample = true;
+ }
+
+ return true;
}
void
-gen6_blorp_hiz_exec(struct brw_context *brw, struct intel_mipmap_tree *mt,
- unsigned int level, unsigned int layer, enum gen6_hiz_op op)
+brw_blorp_resolve_color(struct brw_context *brw, struct intel_mipmap_tree *mt,
+ unsigned level, unsigned layer)
{
- struct brw_blorp_params params;
- brw_blorp_params_init(¶ms);
+ DBG("%s to mt %p level %u layer %u\n", __FUNCTION__, mt, level, layer);
- params.hiz_op = op;
+ const mesa_format format = _mesa_get_srgb_format_linear(mt->format);
- brw_blorp_surface_info_init(brw, ¶ms.depth, mt, level, layer,
- mt->format, true);
+ struct isl_surf isl_tmp[2];
+ struct blorp_surf surf;
+ blorp_surf_for_miptree(brw, &surf, mt, true,
+ (1 << ISL_AUX_USAGE_CCS_E) |
+ (1 << ISL_AUX_USAGE_CCS_D),
+ &level, layer, 1 /* num_layers */,
+ isl_tmp);
- /* Align the rectangle primitive to 8x4 pixels.
- *
- * During fast depth clears, the emitted rectangle primitive must be
- * aligned to 8x4 pixels. From the Ivybridge PRM, Vol 2 Part 1 Section
- * 11.5.3.1 Depth Buffer Clear (and the matching section in the Sandybridge
- * PRM):
- * If Number of Multisamples is NUMSAMPLES_1, the rectangle must be
- * aligned to an 8x4 pixel block relative to the upper left corner
- * of the depth buffer [...]
- *
- * For hiz resolves, the rectangle must also be 8x4 aligned. Item
- * WaHizAmbiguate8x4Aligned from the Haswell workarounds page and the
- * Ivybridge simulator require the alignment.
+ enum blorp_fast_clear_op resolve_op;
+ if (brw->gen >= 9) {
+ if (surf.aux_usage == ISL_AUX_USAGE_CCS_E)
+ resolve_op = BLORP_FAST_CLEAR_OP_RESOLVE_FULL;
+ else
+ resolve_op = BLORP_FAST_CLEAR_OP_RESOLVE_PARTIAL;
+ } else {
+ assert(surf.aux_usage == ISL_AUX_USAGE_CCS_D);
+ /* Broadwell and earlier do not have a partial resolve */
+ resolve_op = BLORP_FAST_CLEAR_OP_RESOLVE_FULL;
+ }
+
+ /* Ivybrigde PRM Vol 2, Part 1, "11.7 MCS Buffer for Render Target(s)":
*
- * To be safe, let's just align the rect for all hiz operations and all
- * hardware generations.
+ * "Any transition from any value in {Clear, Render, Resolve} to a
+ * different value in {Clear, Render, Resolve} requires end of pipe
+ * synchronization."
*
- * However, for some miptree slices of a Z24 texture, emitting an 8x4
- * aligned rectangle that covers the slice may clobber adjacent slices if
- * we strictly adhered to the texture alignments specified in the PRM. The
- * Ivybridge PRM, Section "Alignment Unit Size", states that
- * SURFACE_STATE.Surface_Horizontal_Alignment should be 4 for Z24 surfaces,
- * not 8. But commit 1f112cc increased the alignment from 4 to 8, which
- * prevents the clobbering.
+ * In other words, fast clear ops are not properly synchronized with
+ * other drawing. We need to use a PIPE_CONTROL to ensure that the
+ * contents of the previous draw hit the render target before we resolve
+ * and again afterwards to ensure that the resolve is complete before we
+ * do any more regular drawing.
*/
- params.dst.num_samples = mt->num_samples;
- if (params.dst.num_samples > 1) {
- params.depth.width = ALIGN(mt->logical_width0, 8);
- params.depth.height = ALIGN(mt->logical_height0, 4);
- } else {
- params.depth.width = ALIGN(params.depth.width, 8);
- params.depth.height = ALIGN(params.depth.height, 4);
- }
+ brw_emit_pipe_control_flush(brw,
+ PIPE_CONTROL_RENDER_TARGET_FLUSH |
+ PIPE_CONTROL_CS_STALL);
+
- params.x1 = params.depth.width;
- params.y1 = params.depth.height;
+ struct blorp_batch batch;
+ blorp_batch_init(&brw->blorp, &batch, brw, 0);
+ blorp_ccs_resolve(&batch, &surf, level, layer,
+ brw_blorp_to_isl_format(brw, format, true),
+ resolve_op);
+ blorp_batch_finish(&batch);
+ /* See comment above */
+ brw_emit_pipe_control_flush(brw,
+ PIPE_CONTROL_RENDER_TARGET_FLUSH |
+ PIPE_CONTROL_CS_STALL);
+}
+
+static void
+gen6_blorp_hiz_exec(struct brw_context *brw, struct intel_mipmap_tree *mt,
+ unsigned int level, unsigned int layer, enum blorp_hiz_op op)
+{
assert(intel_miptree_level_has_hiz(mt, level));
- switch (mt->format) {
- case MESA_FORMAT_Z_UNORM16:
- params.depth_format = BRW_DEPTHFORMAT_D16_UNORM;
+ struct isl_surf isl_tmp[2];
+ struct blorp_surf surf;
+ blorp_surf_for_miptree(brw, &surf, mt, true, (1 << ISL_AUX_USAGE_HIZ),
+ &level, layer, 1, isl_tmp);
+
+ struct blorp_batch batch;
+ blorp_batch_init(&brw->blorp, &batch, brw, 0);
+ blorp_gen6_hiz_op(&batch, &surf, level, layer, op);
+ blorp_batch_finish(&batch);
+}
+
+/**
+ * Perform a HiZ or depth resolve operation.
+ *
+ * For an overview of HiZ ops, see the following sections of the Sandy Bridge
+ * PRM, Volume 1, Part 2:
+ * - 7.5.3.1 Depth Buffer Clear
+ * - 7.5.3.2 Depth Buffer Resolve
+ * - 7.5.3.3 Hierarchical Depth Buffer Resolve
+ */
+void
+intel_hiz_exec(struct brw_context *brw, struct intel_mipmap_tree *mt,
+ unsigned int level, unsigned int layer, enum blorp_hiz_op op)
+{
+ const char *opname = NULL;
+
+ switch (op) {
+ case BLORP_HIZ_OP_DEPTH_RESOLVE:
+ opname = "depth resolve";
break;
- case MESA_FORMAT_Z_FLOAT32:
- params.depth_format = BRW_DEPTHFORMAT_D32_FLOAT;
+ case BLORP_HIZ_OP_HIZ_RESOLVE:
+ opname = "hiz ambiguate";
break;
- case MESA_FORMAT_Z24_UNORM_X8_UINT:
- params.depth_format = BRW_DEPTHFORMAT_D24_UNORM_X8_UINT;
+ case BLORP_HIZ_OP_DEPTH_CLEAR:
+ opname = "depth clear";
+ break;
+ case BLORP_HIZ_OP_NONE:
+ opname = "noop?";
break;
- default:
- unreachable("not reached");
}
- brw_blorp_exec(brw, ¶ms);
+ DBG("%s %s to mt %p level %d layer %d\n",
+ __func__, opname, mt, level, layer);
+
+ if (brw->gen >= 8) {
+ gen8_hiz_exec(brw, mt, level, layer, op);
+ } else {
+ gen6_blorp_hiz_exec(brw, mt, level, layer, op);
+ }
}