/**
* Sets up a DRIImage structure to point to our shared image in a region
*/
static void
intel_setup_image_from_mipmap_tree(struct intel_context *intel, __DRIimage *image,
struct intel_mipmap_tree *mt, GLuint level,
GLuint zoffset)
{
unsigned int draw_x, draw_y;
uint32_t mask_x, mask_y;
intel_miptree_check_level_layer(mt, level, zoffset);
intel_region_get_tile_masks(mt->region, &mask_x, &mask_y, false);
intel_miptree_get_image_offset(mt, level, zoffset, &draw_x, &draw_y);
image->width = mt->level[level].width;
image->height = mt->level[level].height;
image->tile_x = draw_x & mask_x;
image->tile_y = draw_y & mask_y;
image->offset = intel_region_get_aligned_offset(mt->region,
draw_x & ~mask_x,
draw_y & ~mask_y,
false);
intel_region_reference(&image->region, mt->region);
}
/**
* Rendering to tiled buffers requires that the base address of the
* buffer be aligned to a page boundary. We generally render to
* textures by pointing the surface at the mipmap image level, which
* may not be aligned to a tile boundary.
*
* This function returns an appropriately-aligned base offset
* according to the tiling restrictions, plus any required x/y offset
* from there.
*/
uint32_t
intel_renderbuffer_tile_offsets(struct intel_renderbuffer *irb,
uint32_t *tile_x,
uint32_t *tile_y)
{
struct intel_region *region = irb->mt->region;
uint32_t mask_x, mask_y;
intel_region_get_tile_masks(region, &mask_x, &mask_y, false);
*tile_x = irb->draw_x & mask_x;
*tile_y = irb->draw_y & mask_y;
return intel_region_get_aligned_offset(region, irb->draw_x & ~mask_x,
irb->draw_y & ~mask_y, false);
}
/**
* 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.
*/
uint32_t
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);
*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);
}
/**
* Rendering with tiled buffers requires that the base address of the buffer
* be aligned to a page boundary. For renderbuffers, and sometimes with
* textures, we may want the surface to point at a texture image level that
* isn't at a page boundary.
*
* This function returns an appropriately-aligned base offset
* according to the tiling restrictions, plus any required x/y offset
* from there.
*/
uint32_t
intel_miptree_get_tile_offsets(struct intel_mipmap_tree *mt,
GLuint level, GLuint slice,
uint32_t *tile_x,
uint32_t *tile_y)
{
struct intel_region *region = mt->region;
uint32_t x, y;
uint32_t mask_x, mask_y;
intel_region_get_tile_masks(region, &mask_x, &mask_y, false);
intel_miptree_get_image_offset(mt, level, slice, &x, &y);
*tile_x = x & mask_x;
*tile_y = y & mask_y;
return intel_region_get_aligned_offset(region, x & ~mask_x, y & ~mask_y,
false);
}
static void
gen6_blorp_emit_depth_stencil_config(struct brw_context *brw,
const brw_blorp_params *params)
{
struct intel_context *intel = &brw->intel;
struct gl_context *ctx = &intel->ctx;
uint32_t draw_x = params->depth.x_offset;
uint32_t draw_y = params->depth.y_offset;
uint32_t tile_mask_x, tile_mask_y;
brw_get_depthstencil_tile_masks(params->depth.mt,
params->depth.level,
params->depth.layer,
NULL,
&tile_mask_x, &tile_mask_y);
/* 3DSTATE_DEPTH_BUFFER */
{
uint32_t tile_x = draw_x & tile_mask_x;
uint32_t tile_y = draw_y & tile_mask_y;
uint32_t offset =
intel_region_get_aligned_offset(params->depth.mt->region,
draw_x & ~tile_mask_x,
draw_y & ~tile_mask_y, false);
/* According to the Sandy Bridge PRM, volume 2 part 1, pp326-327
* (3DSTATE_DEPTH_BUFFER dw5), in the documentation for "Depth
* Coordinate Offset X/Y":
*
* "The 3 LSBs of both offsets must be zero to ensure correct
* alignment"
*
* We have no guarantee that tile_x and tile_y are correctly aligned,
* since they are determined by the mipmap layout, which is only aligned
* to multiples of 4.
*
* So, to avoid hanging the GPU, just smash the low order 3 bits of
* tile_x and tile_y to 0. This is a temporary workaround until we come
* up with a better solution.
*/
WARN_ONCE((tile_x & 7) || (tile_y & 7),
"Depth/stencil buffer needs alignment to 8-pixel boundaries.\n"
"Truncating offset, bad rendering may occur.\n");
tile_x &= ~7;
tile_y &= ~7;
intel_emit_post_sync_nonzero_flush(intel);
intel_emit_depth_stall_flushes(intel);
BEGIN_BATCH(7);
OUT_BATCH(_3DSTATE_DEPTH_BUFFER << 16 | (7 - 2));
OUT_BATCH((params->depth.mt->region->pitch - 1) |
params->depth_format << 18 |
1 << 21 | /* separate stencil enable */
1 << 22 | /* hiz enable */
BRW_TILEWALK_YMAJOR << 26 |
1 << 27 | /* y-tiled */
BRW_SURFACE_2D << 29);
OUT_RELOC(params->depth.mt->region->bo,
I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
offset);
OUT_BATCH(BRW_SURFACE_MIPMAPLAYOUT_BELOW << 1 |
(params->depth.width + tile_x - 1) << 6 |
(params->depth.height + tile_y - 1) << 19);
OUT_BATCH(0);
OUT_BATCH(tile_x |
tile_y << 16);
OUT_BATCH(0);
ADVANCE_BATCH();
}
/* 3DSTATE_HIER_DEPTH_BUFFER */
{
struct intel_region *hiz_region = params->depth.mt->hiz_mt->region;
uint32_t hiz_offset =
intel_region_get_aligned_offset(hiz_region,
draw_x & ~tile_mask_x,
(draw_y & ~tile_mask_y) / 2, false);
BEGIN_BATCH(3);
OUT_BATCH((_3DSTATE_HIER_DEPTH_BUFFER << 16) | (3 - 2));
OUT_BATCH(hiz_region->pitch - 1);
OUT_RELOC(hiz_region->bo,
I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
hiz_offset);
ADVANCE_BATCH();
}
/* 3DSTATE_STENCIL_BUFFER */
{
BEGIN_BATCH(3);
OUT_BATCH((_3DSTATE_STENCIL_BUFFER << 16) | (3 - 2));
OUT_BATCH(0);
OUT_BATCH(0);
ADVANCE_BATCH();
}
}
void
brw_workaround_depthstencil_alignment(struct brw_context *brw)
{
struct intel_context *intel = &brw->intel;
struct gl_context *ctx = &intel->ctx;
struct gl_framebuffer *fb = ctx->DrawBuffer;
bool rebase_depth = false;
bool rebase_stencil = false;
struct intel_renderbuffer *depth_irb = intel_get_renderbuffer(fb, BUFFER_DEPTH);
struct intel_renderbuffer *stencil_irb = intel_get_renderbuffer(fb, BUFFER_STENCIL);
struct intel_mipmap_tree *depth_mt = NULL;
struct intel_mipmap_tree *stencil_mt = get_stencil_miptree(stencil_irb);
uint32_t tile_x = 0, tile_y = 0, stencil_tile_x = 0, stencil_tile_y = 0;
uint32_t stencil_draw_x = 0, stencil_draw_y = 0;
if (depth_irb)
depth_mt = depth_irb->mt;
uint32_t tile_mask_x, tile_mask_y;
brw_get_depthstencil_tile_masks(depth_mt, stencil_mt,
&tile_mask_x, &tile_mask_y);
if (depth_irb) {
tile_x = depth_irb->draw_x & tile_mask_x;
tile_y = depth_irb->draw_y & tile_mask_y;
/* According to the Sandy Bridge PRM, volume 2 part 1, pp326-327
* (3DSTATE_DEPTH_BUFFER dw5), in the documentation for "Depth
* Coordinate Offset X/Y":
*
* "The 3 LSBs of both offsets must be zero to ensure correct
* alignment"
*/
if (tile_x & 7 || tile_y & 7)
rebase_depth = true;
/* We didn't even have intra-tile offsets before g45. */
if (intel->gen == 4 && !intel->is_g4x) {
if (tile_x || tile_y)
rebase_depth = true;
}
if (rebase_depth) {
perf_debug("HW workaround: blitting depth level %d to a temporary "
"to fix alignment (depth tile offset %d,%d)\n",
depth_irb->mt_level, tile_x, tile_y);
intel_renderbuffer_move_to_temp(intel, depth_irb);
/* In the case of stencil_irb being the same packed depth/stencil
* texture but not the same rb, make it point at our rebased mt, too.
*/
if (stencil_irb &&
stencil_irb != depth_irb &&
stencil_irb->mt == depth_mt) {
intel_miptree_reference(&stencil_irb->mt, depth_irb->mt);
intel_renderbuffer_set_draw_offset(stencil_irb);
}
stencil_mt = get_stencil_miptree(stencil_irb);
tile_x = depth_irb->draw_x & tile_mask_x;
tile_y = depth_irb->draw_y & tile_mask_y;
}
if (stencil_irb) {
stencil_mt = get_stencil_miptree(stencil_irb);
intel_miptree_get_image_offset(stencil_mt,
stencil_irb->mt_level,
stencil_irb->mt_layer,
&stencil_draw_x, &stencil_draw_y);
int stencil_tile_x = stencil_draw_x & tile_mask_x;
int stencil_tile_y = stencil_draw_y & tile_mask_y;
/* If stencil doesn't match depth, then we'll need to rebase stencil
* as well. (if we hadn't decided to rebase stencil before, the
* post-stencil depth test will also rebase depth to try to match it
* up).
*/
if (tile_x != stencil_tile_x ||
tile_y != stencil_tile_y) {
rebase_stencil = true;
}
}
}
/* If we have (just) stencil, check it for ignored low bits as well */
if (stencil_irb) {
intel_miptree_get_image_offset(stencil_mt,
stencil_irb->mt_level,
stencil_irb->mt_layer,
&stencil_draw_x, &stencil_draw_y);
stencil_tile_x = stencil_draw_x & tile_mask_x;
stencil_tile_y = stencil_draw_y & tile_mask_y;
if (stencil_tile_x & 7 || stencil_tile_y & 7)
rebase_stencil = true;
if (intel->gen == 4 && !intel->is_g4x) {
if (stencil_tile_x || stencil_tile_y)
rebase_stencil = true;
}
}
if (rebase_stencil) {
perf_debug("HW workaround: blitting stencil level %d to a temporary "
"to fix alignment (stencil tile offset %d,%d)\n",
stencil_irb->mt_level, stencil_tile_x, stencil_tile_y);
intel_renderbuffer_move_to_temp(intel, stencil_irb);
stencil_mt = get_stencil_miptree(stencil_irb);
intel_miptree_get_image_offset(stencil_mt,
stencil_irb->mt_level,
stencil_irb->mt_layer,
&stencil_draw_x, &stencil_draw_y);
stencil_tile_x = stencil_draw_x & tile_mask_x;
stencil_tile_y = stencil_draw_y & tile_mask_y;
if (depth_irb && depth_irb->mt == stencil_irb->mt) {
intel_miptree_reference(&depth_irb->mt, stencil_irb->mt);
intel_renderbuffer_set_draw_offset(depth_irb);
} else if (depth_irb && !rebase_depth) {
if (tile_x != stencil_tile_x ||
tile_y != stencil_tile_y) {
perf_debug("HW workaround: blitting depth level %d to a temporary "
"to match stencil level %d alignment (depth tile offset "
"%d,%d, stencil offset %d,%d)\n",
depth_irb->mt_level,
stencil_irb->mt_level,
tile_x, tile_y,
stencil_tile_x, stencil_tile_y);
intel_renderbuffer_move_to_temp(intel, depth_irb);
tile_x = depth_irb->draw_x & tile_mask_x;
tile_y = depth_irb->draw_y & tile_mask_y;
if (stencil_irb && stencil_irb->mt == depth_mt) {
intel_miptree_reference(&stencil_irb->mt, depth_irb->mt);
intel_renderbuffer_set_draw_offset(stencil_irb);
}
WARN_ONCE(stencil_tile_x != tile_x ||
stencil_tile_y != tile_y,
"Rebased stencil tile offset (%d,%d) doesn't match depth "
"tile offset (%d,%d).\n",
stencil_tile_x, stencil_tile_y,
tile_x, tile_y);
}
}
}
if (!depth_irb) {
tile_x = stencil_tile_x;
tile_y = stencil_tile_y;
}
/* While we just tried to get everything aligned, we may have failed to do
* so in the case of rendering to array or 3D textures, where nonzero faces
* will still have an offset post-rebase. At least give an informative
* warning.
*/
WARN_ONCE((tile_x & 7) || (tile_y & 7),
"Depth/stencil buffer needs alignment to 8-pixel boundaries.\n"
"Truncating offset, bad rendering may occur.\n");
tile_x &= ~7;
tile_y &= ~7;
/* Now, after rebasing, save off the new dephtstencil state so the hardware
* packets can just dereference that without re-calculating tile offsets.
*/
brw->depthstencil.tile_x = tile_x;
brw->depthstencil.tile_y = tile_y;
brw->depthstencil.depth_offset = 0;
brw->depthstencil.stencil_offset = 0;
brw->depthstencil.hiz_offset = 0;
brw->depthstencil.depth_mt = NULL;
brw->depthstencil.stencil_mt = NULL;
brw->depthstencil.hiz_mt = NULL;
if (depth_irb) {
depth_mt = depth_irb->mt;
brw->depthstencil.depth_mt = depth_mt;
brw->depthstencil.depth_offset =
intel_region_get_aligned_offset(depth_mt->region,
depth_irb->draw_x & ~tile_mask_x,
depth_irb->draw_y & ~tile_mask_y,
false);
if (depth_mt->hiz_mt) {
brw->depthstencil.hiz_mt = depth_mt->hiz_mt;
brw->depthstencil.hiz_offset =
intel_region_get_aligned_offset(depth_mt->region,
depth_irb->draw_x & ~tile_mask_x,
(depth_irb->draw_y & ~tile_mask_y) /
2,
false);
}
}
if (stencil_irb) {
stencil_mt = get_stencil_miptree(stencil_irb);
brw->depthstencil.stencil_mt = stencil_mt;
if (stencil_mt->format == MESA_FORMAT_S8) {
/* Note: we can't compute the stencil offset using
* intel_region_get_aligned_offset(), because stencil_region claims
* that the region is untiled even though it's W tiled.
*/
brw->depthstencil.stencil_offset =
(stencil_draw_y & ~tile_mask_y) * stencil_mt->region->pitch +
(stencil_draw_x & ~tile_mask_x) * 64;
}
}
}
void
brw_workaround_depthstencil_alignment(struct brw_context *brw,
GLbitfield clear_mask)
{
struct gl_context *ctx = &brw->ctx;
struct gl_framebuffer *fb = ctx->DrawBuffer;
bool rebase_depth = false;
bool rebase_stencil = false;
struct intel_renderbuffer *depth_irb = intel_get_renderbuffer(fb, BUFFER_DEPTH);
struct intel_renderbuffer *stencil_irb = intel_get_renderbuffer(fb, BUFFER_STENCIL);
struct intel_mipmap_tree *depth_mt = NULL;
struct intel_mipmap_tree *stencil_mt = get_stencil_miptree(stencil_irb);
uint32_t tile_x = 0, tile_y = 0, stencil_tile_x = 0, stencil_tile_y = 0;
uint32_t stencil_draw_x = 0, stencil_draw_y = 0;
bool invalidate_depth = clear_mask & BUFFER_BIT_DEPTH;
bool invalidate_stencil = clear_mask & BUFFER_BIT_STENCIL;
if (depth_irb)
depth_mt = depth_irb->mt;
/* Initialize brw->depthstencil to 'nop' workaround state.
*/
brw->depthstencil.tile_x = 0;
brw->depthstencil.tile_y = 0;
brw->depthstencil.depth_offset = 0;
brw->depthstencil.stencil_offset = 0;
brw->depthstencil.hiz_offset = 0;
brw->depthstencil.depth_mt = NULL;
brw->depthstencil.stencil_mt = NULL;
if (depth_irb)
brw->depthstencil.depth_mt = depth_mt;
if (stencil_irb)
brw->depthstencil.stencil_mt = get_stencil_miptree(stencil_irb);
/* Gen7+ doesn't require the workarounds, since we always program the
* surface state at the start of the whole surface.
*/
if (brw->gen >= 7)
return;
/* Check if depth buffer is in depth/stencil format. If so, then it's only
* safe to invalidate it if we're also clearing stencil, and both depth_irb
* and stencil_irb point to the same miptree.
*
* Note: it's not sufficient to check for the case where
* _mesa_get_format_base_format(depth_mt->format) == GL_DEPTH_STENCIL,
* because this fails to catch depth/stencil buffers on hardware that uses
* separate stencil. To catch that case, we check whether
* depth_mt->stencil_mt is non-NULL.
*/
if (depth_irb && invalidate_depth &&
(_mesa_get_format_base_format(depth_mt->format) == GL_DEPTH_STENCIL ||
depth_mt->stencil_mt)) {
invalidate_depth = invalidate_stencil && depth_irb && stencil_irb
&& depth_irb->mt == stencil_irb->mt;
}
uint32_t tile_mask_x, tile_mask_y;
brw_get_depthstencil_tile_masks(depth_mt,
depth_mt ? depth_irb->mt_level : 0,
depth_mt ? depth_irb->mt_layer : 0,
stencil_mt,
&tile_mask_x, &tile_mask_y);
if (depth_irb) {
tile_x = depth_irb->draw_x & tile_mask_x;
tile_y = depth_irb->draw_y & tile_mask_y;
/* According to the Sandy Bridge PRM, volume 2 part 1, pp326-327
* (3DSTATE_DEPTH_BUFFER dw5), in the documentation for "Depth
* Coordinate Offset X/Y":
*
* "The 3 LSBs of both offsets must be zero to ensure correct
* alignment"
*/
if (tile_x & 7 || tile_y & 7)
rebase_depth = true;
/* We didn't even have intra-tile offsets before g45. */
if (!brw->has_surface_tile_offset) {
if (tile_x || tile_y)
rebase_depth = true;
}
if (rebase_depth) {
perf_debug("HW workaround: blitting depth level %d to a temporary "
"to fix alignment (depth tile offset %d,%d)\n",
depth_irb->mt_level, tile_x, tile_y);
intel_renderbuffer_move_to_temp(brw, depth_irb, invalidate_depth);
/* In the case of stencil_irb being the same packed depth/stencil
* texture but not the same rb, make it point at our rebased mt, too.
*/
if (stencil_irb &&
stencil_irb != depth_irb &&
stencil_irb->mt == depth_mt) {
intel_miptree_reference(&stencil_irb->mt, depth_irb->mt);
intel_renderbuffer_set_draw_offset(stencil_irb);
}
stencil_mt = get_stencil_miptree(stencil_irb);
tile_x = depth_irb->draw_x & tile_mask_x;
tile_y = depth_irb->draw_y & tile_mask_y;
}
if (stencil_irb) {
stencil_mt = get_stencil_miptree(stencil_irb);
intel_miptree_get_image_offset(stencil_mt,
stencil_irb->mt_level,
stencil_irb->mt_layer,
&stencil_draw_x, &stencil_draw_y);
int stencil_tile_x = stencil_draw_x & tile_mask_x;
int stencil_tile_y = stencil_draw_y & tile_mask_y;
/* If stencil doesn't match depth, then we'll need to rebase stencil
* as well. (if we hadn't decided to rebase stencil before, the
* post-stencil depth test will also rebase depth to try to match it
* up).
*/
if (tile_x != stencil_tile_x ||
tile_y != stencil_tile_y) {
rebase_stencil = true;
}
}
}
/* If we have (just) stencil, check it for ignored low bits as well */
if (stencil_irb) {
intel_miptree_get_image_offset(stencil_mt,
stencil_irb->mt_level,
stencil_irb->mt_layer,
&stencil_draw_x, &stencil_draw_y);
stencil_tile_x = stencil_draw_x & tile_mask_x;
stencil_tile_y = stencil_draw_y & tile_mask_y;
if (stencil_tile_x & 7 || stencil_tile_y & 7)
rebase_stencil = true;
if (!brw->has_surface_tile_offset) {
if (stencil_tile_x || stencil_tile_y)
rebase_stencil = true;
}
}
if (rebase_stencil) {
perf_debug("HW workaround: blitting stencil level %d to a temporary "
"to fix alignment (stencil tile offset %d,%d)\n",
stencil_irb->mt_level, stencil_tile_x, stencil_tile_y);
intel_renderbuffer_move_to_temp(brw, stencil_irb, invalidate_stencil);
stencil_mt = get_stencil_miptree(stencil_irb);
intel_miptree_get_image_offset(stencil_mt,
stencil_irb->mt_level,
stencil_irb->mt_layer,
&stencil_draw_x, &stencil_draw_y);
stencil_tile_x = stencil_draw_x & tile_mask_x;
stencil_tile_y = stencil_draw_y & tile_mask_y;
if (depth_irb && depth_irb->mt == stencil_irb->mt) {
intel_miptree_reference(&depth_irb->mt, stencil_irb->mt);
intel_renderbuffer_set_draw_offset(depth_irb);
} else if (depth_irb && !rebase_depth) {
if (tile_x != stencil_tile_x ||
tile_y != stencil_tile_y) {
perf_debug("HW workaround: blitting depth level %d to a temporary "
"to match stencil level %d alignment (depth tile offset "
"%d,%d, stencil offset %d,%d)\n",
depth_irb->mt_level,
stencil_irb->mt_level,
tile_x, tile_y,
stencil_tile_x, stencil_tile_y);
intel_renderbuffer_move_to_temp(brw, depth_irb, invalidate_depth);
tile_x = depth_irb->draw_x & tile_mask_x;
tile_y = depth_irb->draw_y & tile_mask_y;
if (stencil_irb && stencil_irb->mt == depth_mt) {
intel_miptree_reference(&stencil_irb->mt, depth_irb->mt);
intel_renderbuffer_set_draw_offset(stencil_irb);
}
WARN_ONCE(stencil_tile_x != tile_x ||
stencil_tile_y != tile_y,
"Rebased stencil tile offset (%d,%d) doesn't match depth "
"tile offset (%d,%d).\n",
stencil_tile_x, stencil_tile_y,
tile_x, tile_y);
}
}
}
if (!depth_irb) {
tile_x = stencil_tile_x;
tile_y = stencil_tile_y;
}
/* While we just tried to get everything aligned, we may have failed to do
* so in the case of rendering to array or 3D textures, where nonzero faces
* will still have an offset post-rebase. At least give an informative
* warning.
*/
WARN_ONCE((tile_x & 7) || (tile_y & 7),
"Depth/stencil buffer needs alignment to 8-pixel boundaries.\n"
"Truncating offset, bad rendering may occur.\n");
tile_x &= ~7;
tile_y &= ~7;
/* Now, after rebasing, save off the new dephtstencil state so the hardware
* packets can just dereference that without re-calculating tile offsets.
*/
brw->depthstencil.tile_x = tile_x;
brw->depthstencil.tile_y = tile_y;
if (depth_irb) {
depth_mt = depth_irb->mt;
brw->depthstencil.depth_mt = depth_mt;
brw->depthstencil.depth_offset =
intel_region_get_aligned_offset(depth_mt->region,
depth_irb->draw_x & ~tile_mask_x,
depth_irb->draw_y & ~tile_mask_y,
false);
if (intel_renderbuffer_has_hiz(depth_irb)) {
brw->depthstencil.hiz_offset =
intel_region_get_aligned_offset(depth_mt->region,
depth_irb->draw_x & ~tile_mask_x,
(depth_irb->draw_y & ~tile_mask_y) /
2,
false);
}
}
if (stencil_irb) {
stencil_mt = get_stencil_miptree(stencil_irb);
brw->depthstencil.stencil_mt = stencil_mt;
if (stencil_mt->format == MESA_FORMAT_S_UINT8) {
/* Note: we can't compute the stencil offset using
* intel_region_get_aligned_offset(), because stencil_region claims
* that the region is untiled even though it's W tiled.
*/
brw->depthstencil.stencil_offset =
(stencil_draw_y & ~tile_mask_y) * stencil_mt->region->pitch +
(stencil_draw_x & ~tile_mask_x) * 64;
}
}
}
static void
gen7_blorp_emit_depth_stencil_config(struct brw_context *brw,
const brw_blorp_params *params)
{
struct intel_context *intel = &brw->intel;
uint32_t draw_x = params->depth.x_offset;
uint32_t draw_y = params->depth.y_offset;
uint32_t tile_mask_x, tile_mask_y;
gen6_blorp_compute_tile_masks(params, &tile_mask_x, &tile_mask_y);
/* 3DSTATE_DEPTH_BUFFER */
{
uint32_t tile_x = draw_x & tile_mask_x;
uint32_t tile_y = draw_y & tile_mask_y;
uint32_t offset =
intel_region_get_aligned_offset(params->depth.mt->region,
draw_x & ~tile_mask_x,
draw_y & ~tile_mask_y, false);
/* According to the Sandy Bridge PRM, volume 2 part 1, pp326-327
* (3DSTATE_DEPTH_BUFFER dw5), in the documentation for "Depth
* Coordinate Offset X/Y":
*
* "The 3 LSBs of both offsets must be zero to ensure correct
* alignment"
*
* We have no guarantee that tile_x and tile_y are correctly aligned,
* since they are determined by the mipmap layout, which is only aligned
* to multiples of 4.
*
* So, to avoid hanging the GPU, just smash the low order 3 bits of
* tile_x and tile_y to 0. This is a temporary workaround until we come
* up with a better solution.
*/
tile_x &= ~7;
tile_y &= ~7;
intel_emit_depth_stall_flushes(intel);
BEGIN_BATCH(7);
OUT_BATCH(GEN7_3DSTATE_DEPTH_BUFFER << 16 | (7 - 2));
uint32_t pitch_bytes =
params->depth.mt->region->pitch * params->depth.mt->region->cpp;
OUT_BATCH((pitch_bytes - 1) |
params->depth_format << 18 |
1 << 22 | /* hiz enable */
1 << 28 | /* depth write */
BRW_SURFACE_2D << 29);
OUT_RELOC(params->depth.mt->region->bo,
I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
offset);
OUT_BATCH((params->depth.width + tile_x - 1) << 4 |
(params->depth.height + tile_y - 1) << 18);
OUT_BATCH(0);
OUT_BATCH(tile_x |
tile_y << 16);
OUT_BATCH(0);
ADVANCE_BATCH();
}
/* 3DSTATE_HIER_DEPTH_BUFFER */
{
struct intel_region *hiz_region = params->depth.mt->hiz_mt->region;
uint32_t hiz_offset =
intel_region_get_aligned_offset(hiz_region,
draw_x & ~tile_mask_x,
(draw_y & ~tile_mask_y) / 2, false);
BEGIN_BATCH(3);
OUT_BATCH((GEN7_3DSTATE_HIER_DEPTH_BUFFER << 16) | (3 - 2));
OUT_BATCH(hiz_region->pitch * hiz_region->cpp - 1);
OUT_RELOC(hiz_region->bo,
I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
hiz_offset);
ADVANCE_BATCH();
}
/* 3DSTATE_STENCIL_BUFFER */
{
BEGIN_BATCH(3);
OUT_BATCH((GEN7_3DSTATE_STENCIL_BUFFER << 16) | (3 - 2));
OUT_BATCH(0);
OUT_BATCH(0);
ADVANCE_BATCH();
}
}
static void emit_depthbuffer(struct brw_context *brw)
{
struct intel_context *intel = &brw->intel;
struct gl_context *ctx = &intel->ctx;
struct gl_framebuffer *fb = ctx->DrawBuffer;
/* _NEW_BUFFERS */
struct intel_renderbuffer *depth_irb = intel_get_renderbuffer(fb, BUFFER_DEPTH);
struct intel_renderbuffer *stencil_irb = intel_get_renderbuffer(fb, BUFFER_STENCIL);
struct intel_mipmap_tree *stencil_mt = NULL;
struct intel_region *hiz_region = NULL;
unsigned int len;
bool separate_stencil = false;
/* Amount by which drawing should be offset in order to draw to the
* appropriate miplevel/zoffset/cubeface. We will extract these values
* from depth_irb or stencil_irb once we determine which is present.
*/
uint32_t draw_x = 0, draw_y = 0;
/* Masks used to determine how much of the draw_x and draw_y offsets should
* be performed using the fine adjustment of "depth coordinate offset X/Y"
* (dw5 of 3DSTATE_DEPTH_BUFFER). Any remaining coarse adjustment will be
* performed by changing the base addresses of the buffers.
*
* Since the HiZ, depth, and stencil buffers all use the same "depth
* coordinate offset X/Y" values, we need to make sure that the coarse
* adjustment will be possible to apply to all three buffers. Since coarse
* adjustment can only be applied in multiples of the tile size, we will OR
* together the tile masks of all the buffers to determine which offsets to
* perform as fine adjustments.
*/
uint32_t tile_mask_x = 0, tile_mask_y = 0;
if (depth_irb) {
intel_region_get_tile_masks(depth_irb->mt->region,
&tile_mask_x, &tile_mask_y);
}
if (depth_irb &&
depth_irb->mt &&
depth_irb->mt->hiz_mt) {
hiz_region = depth_irb->mt->hiz_mt->region;
uint32_t hiz_tile_mask_x, hiz_tile_mask_y;
intel_region_get_tile_masks(hiz_region,
&hiz_tile_mask_x, &hiz_tile_mask_y);
/* Each HiZ row represents 2 rows of pixels */
hiz_tile_mask_y = hiz_tile_mask_y << 1 | 1;
tile_mask_x |= hiz_tile_mask_x;
tile_mask_y |= hiz_tile_mask_y;
}
/* 3DSTATE_DEPTH_BUFFER, 3DSTATE_STENCIL_BUFFER are both
* non-pipelined state that will need the PIPE_CONTROL workaround.
*/
if (intel->gen == 6) {
intel_emit_post_sync_nonzero_flush(intel);
intel_emit_depth_stall_flushes(intel);
}
/* Find the real separate stencil mt if present. */
if (stencil_irb) {
stencil_mt = stencil_irb->mt;
if (stencil_mt->stencil_mt)
stencil_mt = stencil_mt->stencil_mt;
if (stencil_mt->format == MESA_FORMAT_S8) {
separate_stencil = true;
/* Separate stencil buffer uses 64x64 tiles. */
tile_mask_x |= 63;
tile_mask_y |= 63;
} else {
uint32_t stencil_tile_mask_x, stencil_tile_mask_y;
intel_region_get_tile_masks(stencil_mt->region,
&stencil_tile_mask_x,
&stencil_tile_mask_y);
tile_mask_x |= stencil_tile_mask_x;
tile_mask_y |= stencil_tile_mask_y;
}
}
/* If there's a packed depth/stencil bound to stencil only, we need to
* emit the packed depth/stencil buffer packet.
*/
if (!depth_irb && stencil_irb && !separate_stencil)
depth_irb = stencil_irb;
if (intel->gen >= 6)
len = 7;
else if (intel->is_g4x || intel->gen == 5)
len = 6;
else
len = 5;
if (!depth_irb && !separate_stencil) {
BEGIN_BATCH(len);
OUT_BATCH(_3DSTATE_DEPTH_BUFFER << 16 | (len - 2));
OUT_BATCH((BRW_DEPTHFORMAT_D32_FLOAT << 18) |
(BRW_SURFACE_NULL << 29));
OUT_BATCH(0);
OUT_BATCH(0);
OUT_BATCH(0);
if (intel->is_g4x || intel->gen >= 5)
OUT_BATCH(0);
if (intel->gen >= 6)
OUT_BATCH(0);
ADVANCE_BATCH();
} else if (!depth_irb && separate_stencil) {
uint32_t tile_x, tile_y;
/*
* There exists a separate stencil buffer but no depth buffer.
*
* The stencil buffer inherits most of its fields from
* 3DSTATE_DEPTH_BUFFER: namely the tile walk, surface type, width, and
* height.
*
* Enable the hiz bit because it and the separate stencil bit must have
* the same value. From Section 2.11.5.6.1.1 3DSTATE_DEPTH_BUFFER, Bit
* 1.21 "Separate Stencil Enable":
* [DevIL]: If this field is enabled, Hierarchical Depth Buffer
* Enable must also be enabled.
*
* [DevGT]: This field must be set to the same value (enabled or
* disabled) as Hierarchical Depth Buffer Enable
*
* The tiled bit must be set. From the Sandybridge PRM, Volume 2, Part 1,
* Section 7.5.5.1.1 3DSTATE_DEPTH_BUFFER, Bit 1.27 Tiled Surface:
* [DevGT+]: This field must be set to TRUE.
*/
assert(intel->has_separate_stencil);
draw_x = stencil_irb->draw_x;
draw_y = stencil_irb->draw_y;
tile_x = draw_x & tile_mask_x;
tile_y = draw_y & tile_mask_y;
/* According to the Sandy Bridge PRM, volume 2 part 1, pp326-327
* (3DSTATE_DEPTH_BUFFER dw5), in the documentation for "Depth
* Coordinate Offset X/Y":
*
* "The 3 LSBs of both offsets must be zero to ensure correct
* alignment"
*
* We have no guarantee that tile_x and tile_y are correctly aligned,
* since they are determined by the mipmap layout, which is only aligned
* to multiples of 4.
*
* So, to avoid hanging the GPU, just smash the low order 3 bits of
* tile_x and tile_y to 0. This is a temporary workaround until we come
* up with a better solution.
*/
tile_x &= ~7;
tile_y &= ~7;
BEGIN_BATCH(len);
OUT_BATCH(_3DSTATE_DEPTH_BUFFER << 16 | (len - 2));
OUT_BATCH((BRW_DEPTHFORMAT_D32_FLOAT << 18) |
(1 << 21) | /* separate stencil enable */
(1 << 22) | /* hiz enable */
(BRW_TILEWALK_YMAJOR << 26) |
(1 << 27) | /* tiled surface */
(BRW_SURFACE_2D << 29));
OUT_BATCH(0);
OUT_BATCH(((stencil_irb->Base.Base.Width + tile_x - 1) << 6) |
(stencil_irb->Base.Base.Height + tile_y - 1) << 19);
OUT_BATCH(0);
if (intel->is_g4x || intel->gen >= 5)
OUT_BATCH(tile_x | (tile_y << 16));
else
assert(tile_x == 0 && tile_y == 0);
if (intel->gen >= 6)
OUT_BATCH(0);
ADVANCE_BATCH();
} else {
struct intel_region *region = depth_irb->mt->region;
uint32_t tile_x, tile_y, offset;
/* If using separate stencil, hiz must be enabled. */
assert(!separate_stencil || hiz_region);
assert(intel->gen < 6 || region->tiling == I915_TILING_Y);
assert(!hiz_region || region->tiling == I915_TILING_Y);
draw_x = depth_irb->draw_x;
draw_y = depth_irb->draw_y;
tile_x = draw_x & tile_mask_x;
tile_y = draw_y & tile_mask_y;
/* According to the Sandy Bridge PRM, volume 2 part 1, pp326-327
* (3DSTATE_DEPTH_BUFFER dw5), in the documentation for "Depth
* Coordinate Offset X/Y":
*
* "The 3 LSBs of both offsets must be zero to ensure correct
* alignment"
*
* We have no guarantee that tile_x and tile_y are correctly aligned,
* since they are determined by the mipmap layout, which is only aligned
* to multiples of 4.
*
* So, to avoid hanging the GPU, just smash the low order 3 bits of
* tile_x and tile_y to 0. This is a temporary workaround until we come
* up with a better solution.
*/
tile_x &= ~7;
tile_y &= ~7;
offset = intel_region_get_aligned_offset(region,
draw_x & ~tile_mask_x,
draw_y & ~tile_mask_y);
BEGIN_BATCH(len);
OUT_BATCH(_3DSTATE_DEPTH_BUFFER << 16 | (len - 2));
OUT_BATCH(((region->pitch * region->cpp) - 1) |
(brw_depthbuffer_format(brw) << 18) |
((hiz_region ? 1 : 0) << 21) | /* separate stencil enable */
((hiz_region ? 1 : 0) << 22) | /* hiz enable */
(BRW_TILEWALK_YMAJOR << 26) |
((region->tiling != I915_TILING_NONE) << 27) |
(BRW_SURFACE_2D << 29));
OUT_RELOC(region->bo,
I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
offset);
OUT_BATCH((BRW_SURFACE_MIPMAPLAYOUT_BELOW << 1) |
(((depth_irb->Base.Base.Width + tile_x) - 1) << 6) |
(((depth_irb->Base.Base.Height + tile_y) - 1) << 19));
OUT_BATCH(0);
if (intel->is_g4x || intel->gen >= 5)
OUT_BATCH(tile_x | (tile_y << 16));
else
assert(tile_x == 0 && tile_y == 0);
if (intel->gen >= 6)
OUT_BATCH(0);
ADVANCE_BATCH();
}
if (hiz_region || separate_stencil) {
/*
* In the 3DSTATE_DEPTH_BUFFER batch emitted above, the 'separate
* stencil enable' and 'hiz enable' bits were set. Therefore we must
* emit 3DSTATE_HIER_DEPTH_BUFFER and 3DSTATE_STENCIL_BUFFER. Even if
* there is no stencil buffer, 3DSTATE_STENCIL_BUFFER must be emitted;
* failure to do so causes hangs on gen5 and a stall on gen6.
*/
/* Emit hiz buffer. */
if (hiz_region) {
uint32_t hiz_offset =
intel_region_get_aligned_offset(hiz_region,
draw_x & ~tile_mask_x,
(draw_y & ~tile_mask_y) / 2);
BEGIN_BATCH(3);
OUT_BATCH((_3DSTATE_HIER_DEPTH_BUFFER << 16) | (3 - 2));
OUT_BATCH(hiz_region->pitch * hiz_region->cpp - 1);
OUT_RELOC(hiz_region->bo,
I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
hiz_offset);
ADVANCE_BATCH();
} else {
BEGIN_BATCH(3);
OUT_BATCH((_3DSTATE_HIER_DEPTH_BUFFER << 16) | (3 - 2));
OUT_BATCH(0);
OUT_BATCH(0);
ADVANCE_BATCH();
}
/* Emit stencil buffer. */
if (separate_stencil) {
struct intel_region *region = stencil_mt->region;
/* Note: we can't compute the stencil offset using
* intel_region_get_aligned_offset(), because stencil_region claims
* that the region is untiled; in fact it's W tiled.
*/
uint32_t stencil_offset =
(draw_y & ~tile_mask_y) * region->pitch +
(draw_x & ~tile_mask_x) * 64;
BEGIN_BATCH(3);
OUT_BATCH((_3DSTATE_STENCIL_BUFFER << 16) | (3 - 2));
/* The stencil buffer has quirky pitch requirements. From Vol 2a,
* 11.5.6.2.1 3DSTATE_STENCIL_BUFFER, field "Surface Pitch":
* The pitch must be set to 2x the value computed based on width, as
* the stencil buffer is stored with two rows interleaved.
*/
OUT_BATCH(2 * region->pitch * region->cpp - 1);
OUT_RELOC(region->bo,
I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER,
stencil_offset);
ADVANCE_BATCH();
} else {
BEGIN_BATCH(3);
OUT_BATCH((_3DSTATE_STENCIL_BUFFER << 16) | (3 - 2));
OUT_BATCH(0);
OUT_BATCH(0);
ADVANCE_BATCH();
}
}
/*
* On Gen >= 6, emit clear params for safety. If using hiz, then clear
* params must be emitted.
*
* From Section 2.11.5.6.4.1 3DSTATE_CLEAR_PARAMS:
* 3DSTATE_CLEAR_PARAMS packet must follow the DEPTH_BUFFER_STATE packet
* when HiZ is enabled and the DEPTH_BUFFER_STATE changes.
*/
if (intel->gen >= 6 || hiz_region) {
if (intel->gen == 6)
intel_emit_post_sync_nonzero_flush(intel);
BEGIN_BATCH(2);
OUT_BATCH(_3DSTATE_CLEAR_PARAMS << 16 |
GEN5_DEPTH_CLEAR_VALID |
(2 - 2));
OUT_BATCH(depth_irb ? depth_irb->mt->depth_clear_value : 0);
ADVANCE_BATCH();
}
}