* IN THE SOFTWARE.
*/
+#include <errno.h>
#include "intel_batchbuffer.h"
#include "intel_fbo.h"
#include "brw_blorp.h"
#include "brw_defines.h"
+#include "brw_state.h"
#include "gen6_blorp.h"
#include "gen7_blorp.h"
+#define FILE_DEBUG_FLAG DEBUG_BLORP
+
brw_blorp_mip_info::brw_blorp_mip_info()
: mt(NULL),
level(0),
brw_blorp_mip_info::set(struct intel_mipmap_tree *mt,
unsigned int level, unsigned int layer)
{
+ /* 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(layer % mt->num_samples == 0);
+ }
+
intel_miptree_check_level_layer(mt, level, layer);
this->mt = mt;
this->level = level;
this->layer = layer;
- this->width = mt->level[level].width;
- this->height = mt->level[level].height;
+ this->width = minify(mt->physical_width0, level - mt->first_level);
+ this->height = minify(mt->physical_height0, level - mt->first_level);
intel_miptree_get_image_offset(mt, level, layer, &x_offset, &y_offset);
}
void
brw_blorp_surface_info::set(struct brw_context *brw,
struct intel_mipmap_tree *mt,
- unsigned int level, unsigned int layer)
+ unsigned int level, unsigned int layer,
+ mesa_format format, bool is_render_target)
{
brw_blorp_mip_info::set(mt, level, layer);
this->num_samples = mt->num_samples;
- this->array_spacing_lod0 = mt->array_spacing_lod0;
+ this->array_layout = mt->array_layout;
this->map_stencil_as_y_tiled = false;
this->msaa_layout = mt->msaa_layout;
- switch (mt->format) {
- case MESA_FORMAT_S8:
+ if (format == MESA_FORMAT_NONE)
+ format = mt->format;
+
+ switch (format) {
+ 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.
this->map_stencil_as_y_tiled = true;
this->brw_surfaceformat = BRW_SURFACEFORMAT_R8_UNORM;
break;
- case MESA_FORMAT_X8_Z24:
- case MESA_FORMAT_Z32_FLOAT:
- /* The miptree consists of 32 bits per pixel, arranged either as 24-bit
- * depth values interleaved with 8 "don't care" bits, or as 32-bit
- * floating point depth values. 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.
+ 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.
*/
this->brw_surfaceformat = BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
break;
- case MESA_FORMAT_Z16:
- /* The miptree consists of 16 bits per pixel of depth data. 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 is as 8-bit RG.
- */
- this->brw_surfaceformat = BRW_SURFACEFORMAT_R8G8_UNORM;
+ case MESA_FORMAT_Z_FLOAT32:
+ this->brw_surfaceformat = BRW_SURFACEFORMAT_R32_FLOAT;
break;
- default:
- /* Blorp blits don't support any sort of format conversion (except
- * between sRGB and linear), so we can safely assume that the format is
- * supported as a render target, even if this is the source image. So
- * we can convert to a surface format using brw->render_target_format.
- */
- assert(brw->format_supported_as_render_target[mt->format]);
- this->brw_surfaceformat = brw->render_target_format[mt->format];
+ case MESA_FORMAT_Z_UNORM16:
+ this->brw_surfaceformat = BRW_SURFACEFORMAT_R16_UNORM;
+ break;
+ default: {
+ mesa_format linear_format = _mesa_get_srgb_format_linear(format);
+ if (is_render_target) {
+ assert(brw->format_supported_as_render_target[linear_format]);
+ this->brw_surfaceformat = brw->render_target_format[linear_format];
+ } else {
+ this->brw_surfaceformat = brw_format_for_mesa_format(linear_format);
+ }
break;
}
+ }
}
brw_blorp_surface_info::compute_tile_offsets(uint32_t *tile_x,
uint32_t *tile_y) const
{
- struct intel_region *region = mt->region;
uint32_t mask_x, mask_y;
- intel_region_get_tile_masks(region, &mask_x, &mask_y,
- map_stencil_as_y_tiled);
+ intel_miptree_get_tile_masks(mt, &mask_x, &mask_y, map_stencil_as_y_tiled);
*tile_x = x_offset & mask_x;
*tile_y = y_offset & mask_y;
- return intel_region_get_aligned_offset(region, x_offset & ~mask_x,
- y_offset & ~mask_y,
- map_stencil_as_y_tiled);
+ return intel_miptree_get_aligned_offset(mt, x_offset & ~mask_x,
+ y_offset & ~mask_y,
+ map_stencil_as_y_tiled);
}
-brw_blorp_params::brw_blorp_params()
+brw_blorp_params::brw_blorp_params(unsigned num_varyings,
+ unsigned num_draw_buffers,
+ unsigned num_layers)
: x0(0),
y0(0),
x1(0),
y1(0),
depth_format(0),
hiz_op(GEN6_HIZ_OP_NONE),
- fast_clear_op(GEN7_FAST_CLEAR_OP_NONE),
- num_samples(0),
- use_wm_prog(false)
+ use_wm_prog(false),
+ num_varyings(num_varyings),
+ num_draw_buffers(num_draw_buffers),
+ num_layers(num_layers)
{
- color_write_disable[0] = false;
- color_write_disable[1] = false;
- color_write_disable[2] = false;
- color_write_disable[3] = false;
}
extern "C" {
void
-intel_hiz_exec(struct intel_context *intel, struct intel_mipmap_tree *mt,
+intel_hiz_exec(struct brw_context *brw, struct intel_mipmap_tree *mt,
unsigned int level, unsigned int layer, gen6_hiz_op op)
{
- brw_hiz_op_params params(mt, level, layer, op);
- brw_blorp_exec(intel, ¶ms);
+ const char *opname = NULL;
+
+ 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;
+ }
+
+ 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 {
+ brw_hiz_op_params params(mt, level, layer, op);
+ brw_blorp_exec(brw, ¶ms);
+ }
}
} /* extern "C" */
void
-brw_blorp_exec(struct intel_context *intel, const brw_blorp_params *params)
+brw_blorp_exec(struct brw_context *brw, const brw_blorp_params *params)
{
- struct brw_context *brw = brw_context(&intel->ctx);
+ struct gl_context *ctx = &brw->ctx;
+ uint32_t estimated_max_batch_usage = 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.
+ */
+ intel_batchbuffer_emit_mi_flush(brw);
- switch (intel->gen) {
+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 = brw->batch.used;
+ uint32_t saved_state_batch_offset = brw->batch.state_batch_offset;
+
+ switch (brw->gen) {
case 6:
- gen6_blorp_exec(intel, params);
+ gen6_blorp_exec(brw, params);
break;
case 7:
- gen7_blorp_exec(intel, params);
+ gen7_blorp_exec(brw, params);
break;
default:
/* BLORP is not supported before Gen6. */
- assert(false);
- break;
+ 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.
+ */
+ assert(brw->batch.bo == saved_bo);
+ assert((brw->batch.used - 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 (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 (unlikely(intel->always_flush_batch))
- intel_batchbuffer_flush(intel);
+ if (unlikely(brw->always_flush_batch))
+ intel_batchbuffer_flush(brw);
/* We've smashed all state compared to what the normal 3D pipeline
* rendering tracks for GL.
*/
- brw->state.dirty.brw = ~0;
- brw->state.dirty.cache = ~0;
- brw->state_batch_count = 0;
- intel->batch.need_workaround_flush = true;
+ brw->ctx.NewDriverState = ~0ull;
+ brw->no_depth_or_stencil = false;
+ brw->ib.type = -1;
/* Flush the sampler cache so any texturing from the destination is
* coherent.
*/
- intel_batchbuffer_emit_mi_flush(intel);
+ intel_batchbuffer_emit_mi_flush(brw);
}
brw_hiz_op_params::brw_hiz_op_params(struct intel_mipmap_tree *mt,
x1 = depth.width;
y1 = depth.height;
- assert(intel_miptree_slice_has_hiz(mt, level, layer));
+ assert(intel_miptree_level_has_hiz(mt, level));
switch (mt->format) {
- case MESA_FORMAT_Z16: depth_format = BRW_DEPTHFORMAT_D16_UNORM; break;
- case MESA_FORMAT_Z32_FLOAT: depth_format = BRW_DEPTHFORMAT_D32_FLOAT; break;
- case MESA_FORMAT_X8_Z24: depth_format = BRW_DEPTHFORMAT_D24_UNORM_X8_UINT; break;
- default: assert(0); break;
+ case MESA_FORMAT_Z_UNORM16: depth_format = BRW_DEPTHFORMAT_D16_UNORM; break;
+ case MESA_FORMAT_Z_FLOAT32: depth_format = BRW_DEPTHFORMAT_D32_FLOAT; break;
+ case MESA_FORMAT_Z24_UNORM_X8_UINT: depth_format = BRW_DEPTHFORMAT_D24_UNORM_X8_UINT; break;
+ default: unreachable("not reached");
}
}