static void r300_blit(struct pipe_context *pipe,
const struct pipe_blit_info *blit_info)
{
struct r300_context *r300 = r300_context(pipe);
struct pipe_framebuffer_state *fb =
(struct pipe_framebuffer_state*)r300->fb_state.state;
struct pipe_blit_info info = *blit_info;
/* Decompress ZMASK. */
if (r300->zmask_in_use && !r300->locked_zbuffer) {
if (fb->zsbuf->texture == info.src.resource ||
fb->zsbuf->texture == info.dst.resource) {
r300_decompress_zmask(r300);
}
}
/* Blit a combined depth-stencil resource as color.
* S8Z24 is the only supported stencil format. */
if ((info.mask & PIPE_MASK_S) &&
info.src.format == PIPE_FORMAT_S8_UINT_Z24_UNORM &&
info.dst.format == PIPE_FORMAT_S8_UINT_Z24_UNORM) {
info.src.format = PIPE_FORMAT_B8G8R8A8_UNORM;
info.dst.format = PIPE_FORMAT_B8G8R8A8_UNORM;
if (info.mask & PIPE_MASK_Z) {
info.mask = PIPE_MASK_RGBA; /* depth+stencil */
} else {
info.mask = PIPE_MASK_B; /* stencil only */
}
}
r300_blitter_begin(r300, R300_BLIT);
util_blitter_blit(r300->blitter, &info);
r300_blitter_end(r300);
}
/* Flush a depth stencil buffer. */
void r300_flush_depth_stencil(struct pipe_context *pipe,
struct pipe_resource *dst,
unsigned level,
unsigned layer)
{
struct r300_context *r300 = r300_context(pipe);
struct pipe_surface *dstsurf, surf_tmpl;
struct r300_texture *tex = r300_texture(dst);
if (!tex->zmask_mem[level])
return;
if (!tex->zmask_in_use[level])
return;
surf_tmpl.format = dst->format;
surf_tmpl.usage = PIPE_BIND_DEPTH_STENCIL;
surf_tmpl.u.tex.level = level;
surf_tmpl.u.tex.first_layer = layer;
surf_tmpl.u.tex.last_layer = layer;
dstsurf = pipe->create_surface(pipe, dst, &surf_tmpl);
r300->z_decomp_rd = TRUE;
r300_blitter_begin(r300, R300_CLEAR_SURFACE);
util_blitter_flush_depth_stencil(r300->blitter, dstsurf);
r300_blitter_end(r300);
r300->z_decomp_rd = FALSE;
tex->zmask_in_use[level] = FALSE;
}
/* Clear a region of a depth stencil surface. */
static void r300_clear_depth_stencil(struct pipe_context *pipe,
struct pipe_surface *dst,
unsigned clear_flags,
double depth,
unsigned stencil,
unsigned dstx, unsigned dsty,
unsigned width, unsigned height,
bool render_condition_enabled)
{
struct r300_context *r300 = r300_context(pipe);
struct pipe_framebuffer_state *fb =
(struct pipe_framebuffer_state*)r300->fb_state.state;
if (r300->zmask_in_use && !r300->locked_zbuffer) {
if (fb->zsbuf->texture == dst->texture) {
r300_decompress_zmask(r300);
}
}
/* XXX Do not decompress ZMask of the currently-set zbuffer. */
r300_blitter_begin(r300, R300_CLEAR_SURFACE |
(render_condition_enabled ? 0 : R300_IGNORE_RENDER_COND));
util_blitter_clear_depth_stencil(r300->blitter, dst, clear_flags, depth, stencil,
dstx, dsty, width, height);
r300_blitter_end(r300);
}
static void r300_blit(struct pipe_context *pipe,
const struct pipe_blit_info *blit)
{
struct r300_context *r300 = r300_context(pipe);
struct pipe_framebuffer_state *fb =
(struct pipe_framebuffer_state*)r300->fb_state.state;
struct pipe_blit_info info = *blit;
/* MSAA resolve. */
if (info.src.resource->nr_samples > 1 &&
!util_format_is_depth_or_stencil(info.src.resource->format)) {
r300_msaa_resolve(pipe, &info);
return;
}
/* Can't read MSAA textures. */
if (info.src.resource->nr_samples > 1) {
return;
}
/* Blit a combined depth-stencil resource as color.
* S8Z24 is the only supported stencil format. */
if ((info.mask & PIPE_MASK_S) &&
info.src.format == PIPE_FORMAT_S8_UINT_Z24_UNORM &&
info.dst.format == PIPE_FORMAT_S8_UINT_Z24_UNORM) {
if (info.dst.resource->nr_samples > 1) {
/* Cannot do that with MSAA buffers. */
info.mask &= ~PIPE_MASK_S;
if (!(info.mask & PIPE_MASK_Z)) {
return;
}
} else {
/* Single-sample buffer. */
info.src.format = PIPE_FORMAT_B8G8R8A8_UNORM;
info.dst.format = PIPE_FORMAT_B8G8R8A8_UNORM;
if (info.mask & PIPE_MASK_Z) {
info.mask = PIPE_MASK_RGBA; /* depth+stencil */
} else {
info.mask = PIPE_MASK_B; /* stencil only */
}
}
}
/* Decompress ZMASK. */
if (r300->zmask_in_use && !r300->locked_zbuffer) {
if (fb->zsbuf->texture == info.src.resource ||
fb->zsbuf->texture == info.dst.resource) {
r300_decompress_zmask(r300);
}
}
r300_blitter_begin(r300, R300_BLIT |
(info.render_condition_enable ? 0 : R300_IGNORE_RENDER_COND));
util_blitter_blit(r300->blitter, &info);
r300_blitter_end(r300);
}
/* Clear a region of a color surface to a constant value. */
static void r300_clear_render_target(struct pipe_context *pipe,
struct pipe_surface *dst,
const union pipe_color_union *color,
unsigned dstx, unsigned dsty,
unsigned width, unsigned height)
{
struct r300_context *r300 = r300_context(pipe);
r300_blitter_begin(r300, R300_CLEAR_SURFACE);
util_blitter_clear_render_target(r300->blitter, dst, color,
dstx, dsty, width, height);
r300_blitter_end(r300);
}
/* Copy a block of pixels from one surface to another using HW. */
static void r300_hw_copy_region(struct pipe_context* pipe,
struct pipe_resource *dst,
unsigned dst_level,
unsigned dstx, unsigned dsty, unsigned dstz,
struct pipe_resource *src,
unsigned src_level,
const struct pipe_box *src_box)
{
struct r300_context* r300 = r300_context(pipe);
r300_blitter_begin(r300, R300_COPY);
util_blitter_copy_texture(r300->blitter, dst, dst_level, dstx, dsty, dstz,
src, src_level, src_box, TRUE);
r300_blitter_end(r300);
}
/* Clear a region of a depth stencil surface. */
static void r300_clear_depth_stencil(struct pipe_context *pipe,
struct pipe_surface *dst,
unsigned clear_flags,
double depth,
unsigned stencil,
unsigned dstx, unsigned dsty,
unsigned width, unsigned height)
{
struct r300_context *r300 = r300_context(pipe);
r300_blitter_begin(r300, R300_CLEAR_SURFACE);
util_blitter_clear_depth_stencil(r300->blitter, dst, clear_flags, depth, stencil,
dstx, dsty, width, height);
r300_blitter_end(r300);
}
/* Clear a region of a color surface to a constant value. */
static void r300_clear_render_target(struct pipe_context *pipe,
struct pipe_surface *dst,
const union pipe_color_union *color,
unsigned dstx, unsigned dsty,
unsigned width, unsigned height,
bool render_condition_enabled)
{
struct r300_context *r300 = r300_context(pipe);
r300_blitter_begin(r300, R300_CLEAR_SURFACE |
(render_condition_enabled ? 0 : R300_IGNORE_RENDER_COND));
util_blitter_clear_render_target(r300->blitter, dst, color,
dstx, dsty, width, height);
r300_blitter_end(r300);
}
static void r300_msaa_resolve(struct pipe_context *pipe,
const struct pipe_blit_info *info)
{
struct r300_context *r300 = r300_context(pipe);
struct pipe_screen *screen = pipe->screen;
struct pipe_resource *tmp, templ;
struct pipe_blit_info blit;
assert(info->src.level == 0);
assert(info->src.box.z == 0);
assert(info->src.box.depth == 1);
assert(info->dst.box.depth == 1);
if (r300_is_simple_msaa_resolve(info)) {
r300_simple_msaa_resolve(pipe, info->dst.resource, info->dst.level,
info->dst.box.z, info->src.resource,
info->src.format);
return;
}
/* resolve into a temporary texture, then blit */
memset(&templ, 0, sizeof(templ));
templ.target = PIPE_TEXTURE_2D;
templ.format = info->src.resource->format;
templ.width0 = info->src.resource->width0;
templ.height0 = info->src.resource->height0;
templ.depth0 = 1;
templ.array_size = 1;
templ.usage = PIPE_USAGE_DEFAULT;
templ.flags = R300_RESOURCE_FORCE_MICROTILING;
tmp = screen->resource_create(screen, &templ);
/* resolve */
r300_simple_msaa_resolve(pipe, tmp, 0, 0, info->src.resource,
info->src.format);
/* blit */
blit = *info;
blit.src.resource = tmp;
blit.src.box.z = 0;
r300_blitter_begin(r300, R300_BLIT | R300_IGNORE_RENDER_COND);
util_blitter_blit(r300->blitter, &blit);
r300_blitter_end(r300);
pipe_resource_reference(&tmp, NULL);
}
static void r300_simple_msaa_resolve(struct pipe_context *pipe,
struct pipe_resource *dst,
unsigned dst_level,
unsigned dst_layer,
struct pipe_resource *src,
enum pipe_format format)
{
struct r300_context *r300 = r300_context(pipe);
struct r300_surface *srcsurf, *dstsurf;
struct pipe_surface surf_tmpl;
struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state;
memset(&surf_tmpl, 0, sizeof(surf_tmpl));
surf_tmpl.format = format;
srcsurf = r300_surface(pipe->create_surface(pipe, src, &surf_tmpl));
surf_tmpl.format = format;
surf_tmpl.u.tex.level = dst_level;
surf_tmpl.u.tex.first_layer =
surf_tmpl.u.tex.last_layer = dst_layer;
dstsurf = r300_surface(pipe->create_surface(pipe, dst, &surf_tmpl));
/* COLORPITCH should contain the tiling info of the resolve buffer.
* The tiling of the AA buffer isn't programmable anyway. */
srcsurf->pitch &= ~(R300_COLOR_TILE(1) | R300_COLOR_MICROTILE(3));
srcsurf->pitch |= dstsurf->pitch & (R300_COLOR_TILE(1) | R300_COLOR_MICROTILE(3));
/* Enable AA resolve. */
aa->dest = dstsurf;
r300->aa_state.size = 8;
r300_mark_atom_dirty(r300, &r300->aa_state);
/* Resolve the surface. */
r300_blitter_begin(r300, R300_CLEAR_SURFACE);
util_blitter_custom_color(r300->blitter, &srcsurf->base, NULL);
r300_blitter_end(r300);
/* Disable AA resolve. */
aa->dest = NULL;
r300->aa_state.size = 4;
r300_mark_atom_dirty(r300, &r300->aa_state);
pipe_surface_reference((struct pipe_surface**)&srcsurf, NULL);
pipe_surface_reference((struct pipe_surface**)&dstsurf, NULL);
}
void r300_decompress_zmask(struct r300_context *r300)
{
struct pipe_framebuffer_state *fb =
(struct pipe_framebuffer_state*)r300->fb_state.state;
if (!r300->zmask_in_use || r300->locked_zbuffer)
return;
r300->zmask_decompress = TRUE;
r300_mark_atom_dirty(r300, &r300->hyperz_state);
r300_blitter_begin(r300, R300_DECOMPRESS);
util_blitter_clear_depth_custom(r300->blitter, fb->width, fb->height, 0,
r300->dsa_decompress_zmask);
r300_blitter_end(r300);
r300->zmask_decompress = FALSE;
r300->zmask_in_use = FALSE;
r300_mark_atom_dirty(r300, &r300->hyperz_state);
}
/* Clear currently bound buffers. */
static void r300_clear(struct pipe_context* pipe,
unsigned buffers,
const union pipe_color_union *color,
double depth,
unsigned stencil)
{
/* My notes about Zbuffer compression:
*
* 1) The zbuffer must be micro-tiled and whole microtiles must be
* written if compression is enabled. If microtiling is disabled,
* it locks up.
*
* 2) There is ZMASK RAM which contains a compressed zbuffer.
* Each dword of the Z Mask contains compression information
* for 16 4x4 pixel tiles, that is 2 bits for each tile.
* On chips with 2 Z pipes, every other dword maps to a different
* pipe. On newer chipsets, there is a new compression mode
* with 8x8 pixel tiles per 2 bits.
*
* 3) The FASTFILL bit has nothing to do with filling. It only tells hw
* it should look in the ZMASK RAM first before fetching from a real
* zbuffer.
*
* 4) If a pixel is in a cleared state, ZB_DEPTHCLEARVALUE is returned
* during zbuffer reads instead of the value that is actually stored
* in the zbuffer memory. A pixel is in a cleared state when its ZMASK
* is equal to 0. Therefore, if you clear ZMASK with zeros, you may
* leave the zbuffer memory uninitialized, but then you must enable
* compression, so that the ZMASK RAM is actually used.
*
* 5) Each 4x4 (or 8x8) tile is automatically decompressed and recompressed
* during zbuffer updates. A special decompressing operation should be
* used to fully decompress a zbuffer, which basically just stores all
* compressed tiles in ZMASK to the zbuffer memory.
*
* 6) For a 16-bit zbuffer, compression causes a hung with one or
* two samples and should not be used.
*
* 7) FORCE_COMPRESSED_STENCIL_VALUE should be enabled for stencil clears
* to avoid needless decompression.
*
* 8) Fastfill must not be used if reading of compressed Z data is disabled
* and writing of compressed Z data is enabled (RD/WR_COMP_ENABLE),
* i.e. it cannot be used to compress the zbuffer.
*
* 9) ZB_CB_CLEAR does not interact with zbuffer compression in any way.
*
* - Marek
*/
struct r300_context* r300 = r300_context(pipe);
struct pipe_framebuffer_state *fb =
(struct pipe_framebuffer_state*)r300->fb_state.state;
struct r300_hyperz_state *hyperz =
(struct r300_hyperz_state*)r300->hyperz_state.state;
uint32_t width = fb->width;
uint32_t height = fb->height;
uint32_t hyperz_dcv = hyperz->zb_depthclearvalue;
/* Enable fast Z clear.
* The zbuffer must be in micro-tiled mode, otherwise it locks up. */
if (buffers & PIPE_CLEAR_DEPTHSTENCIL) {
boolean zmask_clear, hiz_clear;
zmask_clear = r300_fast_zclear_allowed(r300);
hiz_clear = r300_hiz_clear_allowed(r300);
/* If we need Hyper-Z. */
if (zmask_clear || hiz_clear) {
r300->num_z_clears++;
/* Try to obtain the access to Hyper-Z buffers if we don't have one. */
if (!r300->hyperz_enabled) {
r300->hyperz_enabled =
r300->rws->cs_request_feature(r300->cs,
RADEON_FID_R300_HYPERZ_ACCESS,
TRUE);
if (r300->hyperz_enabled) {
/* Need to emit HyperZ buffer regs for the first time. */
r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
}
}
/* Setup Hyper-Z clears. */
if (r300->hyperz_enabled) {
DBG(r300, DBG_HYPERZ, "r300: Clear memory: %s%s\n",
zmask_clear ? "ZMASK " : "", hiz_clear ? "HIZ" : "");
if (zmask_clear) {
hyperz_dcv = hyperz->zb_depthclearvalue =
r300_depth_clear_value(fb->zsbuf->format, depth, stencil);
r300_mark_atom_dirty(r300, &r300->zmask_clear);
buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
}
if (hiz_clear) {
r300->hiz_clear_value = r300_hiz_clear_value(depth);
r300_mark_atom_dirty(r300, &r300->hiz_clear);
}
}
}
}
/* Enable CBZB clear. */
if (r300_cbzb_clear_allowed(r300, buffers)) {
struct r300_surface *surf = r300_surface(fb->cbufs[0]);
hyperz->zb_depthclearvalue =
r300_depth_clear_cb_value(surf->base.format, color->f);
width = surf->cbzb_width;
height = surf->cbzb_height;
r300->cbzb_clear = TRUE;
r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
}
/* Clear. */
if (buffers) {
enum pipe_format cformat = fb->nr_cbufs ? fb->cbufs[0]->format : PIPE_FORMAT_NONE;
/* Clear using the blitter. */
r300_blitter_begin(r300, R300_CLEAR);
util_blitter_clear(r300->blitter,
width,
height,
fb->nr_cbufs,
buffers, cformat, color, depth, stencil);
r300_blitter_end(r300);
} else if (r300->zmask_clear.dirty || r300->hiz_clear.dirty) {
/* Just clear zmask and hiz now, this does not use the standard draw
* procedure. */
/* Calculate zmask_clear and hiz_clear atom sizes. */
unsigned dwords =
(r300->zmask_clear.dirty ? r300->zmask_clear.size : 0) +
(r300->hiz_clear.dirty ? r300->hiz_clear.size : 0) +
r300_get_num_cs_end_dwords(r300);
/* Reserve CS space. */
if (dwords > (RADEON_MAX_CMDBUF_DWORDS - r300->cs->cdw)) {
r300_flush(&r300->context, RADEON_FLUSH_ASYNC, NULL);
}
/* Emit clear packets. */
if (r300->zmask_clear.dirty) {
r300_emit_zmask_clear(r300, r300->zmask_clear.size,
r300->zmask_clear.state);
r300->zmask_clear.dirty = FALSE;
}
if (r300->hiz_clear.dirty) {
r300_emit_hiz_clear(r300, r300->hiz_clear.size,
r300->hiz_clear.state);
r300->hiz_clear.dirty = FALSE;
}
} else {
assert(0);
}
/* Disable CBZB clear. */
if (r300->cbzb_clear) {
r300->cbzb_clear = FALSE;
hyperz->zb_depthclearvalue = hyperz_dcv;
r300_mark_fb_state_dirty(r300, R300_CHANGED_HYPERZ_FLAG);
}
/* Enable fastfill and/or hiz.
*
* If we cleared zmask/hiz, it's in use now. The Hyper-Z state update
* looks if zmask/hiz is in use and programs hardware accordingly. */
if (r300->zmask_in_use || r300->hiz_in_use) {
r300_mark_atom_dirty(r300, &r300->hyperz_state);
}
}
/* Copy a block of pixels from one surface to another. */
static void r300_resource_copy_region(struct pipe_context *pipe,
struct pipe_resource *dst,
unsigned dst_level,
unsigned dstx, unsigned dsty, unsigned dstz,
struct pipe_resource *src,
unsigned src_level,
const struct pipe_box *src_box)
{
struct pipe_screen *screen = pipe->screen;
struct r300_context *r300 = r300_context(pipe);
struct pipe_framebuffer_state *fb =
(struct pipe_framebuffer_state*)r300->fb_state.state;
unsigned src_width0 = r300_resource(src)->tex.width0;
unsigned src_height0 = r300_resource(src)->tex.height0;
unsigned dst_width0 = r300_resource(dst)->tex.width0;
unsigned dst_height0 = r300_resource(dst)->tex.height0;
unsigned layout;
struct pipe_box box;
struct pipe_sampler_view src_templ, *src_view;
struct pipe_surface dst_templ, *dst_view;
/* Fallback for buffers. */
if ((dst->target == PIPE_BUFFER && src->target == PIPE_BUFFER) ||
!r300_is_blit_supported(dst->format)) {
util_resource_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
src, src_level, src_box);
return;
}
/* The code below changes the texture format so that the copy can be done
* on hardware. E.g. depth-stencil surfaces are copied as RGBA
* colorbuffers. */
util_blitter_default_dst_texture(&dst_templ, dst, dst_level, dstz, src_box);
util_blitter_default_src_texture(&src_templ, src, src_level);
layout = util_format_description(dst_templ.format)->layout;
/* Handle non-renderable plain formats. */
if (layout == UTIL_FORMAT_LAYOUT_PLAIN &&
(!screen->is_format_supported(screen, src_templ.format, src->target,
src->nr_samples,
PIPE_BIND_SAMPLER_VIEW) ||
!screen->is_format_supported(screen, dst_templ.format, dst->target,
dst->nr_samples,
PIPE_BIND_RENDER_TARGET))) {
switch (util_format_get_blocksize(dst_templ.format)) {
case 1:
dst_templ.format = PIPE_FORMAT_I8_UNORM;
break;
case 2:
dst_templ.format = PIPE_FORMAT_B4G4R4A4_UNORM;
break;
case 4:
dst_templ.format = PIPE_FORMAT_B8G8R8A8_UNORM;
break;
case 8:
dst_templ.format = PIPE_FORMAT_R16G16B16A16_UNORM;
break;
default:
debug_printf("r300: copy_region: Unhandled format: %s. Falling back to software.\n"
"r300: copy_region: Software fallback doesn't work for tiled textures.\n",
util_format_short_name(dst_templ.format));
}
src_templ.format = dst_templ.format;
}
/* Handle compressed formats. */
if (layout == UTIL_FORMAT_LAYOUT_S3TC ||
layout == UTIL_FORMAT_LAYOUT_RGTC) {
assert(src_templ.format == dst_templ.format);
box = *src_box;
src_box = &box;
dst_width0 = align(dst_width0, 4);
dst_height0 = align(dst_height0, 4);
src_width0 = align(src_width0, 4);
src_height0 = align(src_height0, 4);
box.width = align(box.width, 4);
box.height = align(box.height, 4);
switch (util_format_get_blocksize(dst_templ.format)) {
case 8:
/* one 4x4 pixel block has 8 bytes.
* we set 1 pixel = 4 bytes ===> 1 block corrensponds to 2 pixels. */
dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
dst_width0 = dst_width0 / 2;
src_width0 = src_width0 / 2;
dstx /= 2;
box.x /= 2;
box.width /= 2;
break;
case 16:
/* one 4x4 pixel block has 16 bytes.
* we set 1 pixel = 4 bytes ===> 1 block corresponds to 4 pixels. */
dst_templ.format = PIPE_FORMAT_R8G8B8A8_UNORM;
break;
}
src_templ.format = dst_templ.format;
dst_height0 = dst_height0 / 4;
src_height0 = src_height0 / 4;
dsty /= 4;
box.y /= 4;
box.height /= 4;
}
/* Fallback for textures. */
if (!screen->is_format_supported(screen, dst_templ.format,
dst->target, dst->nr_samples,
PIPE_BIND_RENDER_TARGET) ||
!screen->is_format_supported(screen, src_templ.format,
src->target, src->nr_samples,
PIPE_BIND_SAMPLER_VIEW)) {
assert(0 && "this shouldn't happen, update r300_is_blit_supported");
util_resource_copy_region(pipe, dst, dst_level, dstx, dsty, dstz,
src, src_level, src_box);
return;
}
/* Decompress ZMASK. */
if (r300->zmask_in_use && !r300->locked_zbuffer) {
if (fb->zsbuf->texture == src ||
fb->zsbuf->texture == dst) {
r300_decompress_zmask(r300);
}
}
dst_view = r300_create_surface_custom(pipe, dst, &dst_templ, dst_width0, dst_height0);
src_view = r300_create_sampler_view_custom(pipe, src, &src_templ, src_width0, src_height0);
r300_blitter_begin(r300, R300_COPY);
util_blitter_blit_generic(r300->blitter, dst_view, dstx, dsty,
abs(src_box->width), abs(src_box->height),
src_view, src_box,
src_width0, src_height0, PIPE_MASK_RGBAZS,
PIPE_TEX_FILTER_NEAREST, NULL, FALSE);
r300_blitter_end(r300);
pipe_surface_reference(&dst_view, NULL);
pipe_sampler_view_reference(&src_view, NULL);
}
/* Clear currently bound buffers. */
static void r300_clear(struct pipe_context* pipe,
unsigned buffers,
const float* rgba,
double depth,
unsigned stencil)
{
/* My notes about fastfill:
*
* 1) Only the zbuffer is cleared.
*
* 2) The zbuffer must be micro-tiled and whole microtiles must be
* written. If microtiling is disabled, it locks up.
*
* 3) There is Z Mask RAM which contains a compressed zbuffer and
* it interacts with fastfill. We should figure out how to use it
* to get more performance.
* This is what we know about the Z Mask:
*
* Each dword of the Z Mask contains compression information
* for 16 4x4 pixel blocks, that is 2 bits for each block.
* On chips with 2 Z pipes, every other dword maps to a different
* pipe.
*
* 4) ZB_DEPTHCLEARVALUE is used to clear the zbuffer and the Z Mask must
* be equal to 0. (clear the Z Mask RAM with zeros)
*
* 5) For 16-bit zbuffer, compression causes a hung with one or
* two samples and should not be used.
*
* 6) FORCE_COMPRESSED_STENCIL_VALUE should be enabled for stencil clears
* to avoid needless decompression.
*
* 7) Fastfill must not be used if reading of compressed Z data is disabled
* and writing of compressed Z data is enabled (RD/WR_COMP_ENABLE),
* i.e. it cannot be used to compress the zbuffer.
*
* 8) ZB_CB_CLEAR does not interact with fastfill in any way.
*
* - Marek
*/
struct r300_context* r300 = r300_context(pipe);
struct pipe_framebuffer_state *fb =
(struct pipe_framebuffer_state*)r300->fb_state.state;
struct r300_hyperz_state *hyperz =
(struct r300_hyperz_state*)r300->hyperz_state.state;
struct r300_texture *zstex =
fb->zsbuf ? r300_texture(fb->zsbuf->texture) : NULL;
uint32_t width = fb->width;
uint32_t height = fb->height;
boolean can_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
uint32_t hyperz_dcv = hyperz->zb_depthclearvalue;
/* Enable fast Z clear.
* The zbuffer must be in micro-tiled mode, otherwise it locks up. */
if ((buffers & PIPE_CLEAR_DEPTHSTENCIL) && can_hyperz) {
hyperz_dcv = hyperz->zb_depthclearvalue =
r300_depth_clear_value(fb->zsbuf->format, depth, stencil);
r300_mark_fb_state_dirty(r300, R300_CHANGED_ZCLEAR_FLAG);
if (zstex->zmask_mem[fb->zsbuf->u.tex.level]) {
r300_mark_atom_dirty(r300, &r300->zmask_clear);
buffers &= ~PIPE_CLEAR_DEPTHSTENCIL;
}
if (zstex->hiz_mem[fb->zsbuf->u.tex.level])
r300_mark_atom_dirty(r300, &r300->hiz_clear);
}
/* Enable CBZB clear. */
if (r300_cbzb_clear_allowed(r300, buffers)) {
struct r300_surface *surf = r300_surface(fb->cbufs[0]);
hyperz->zb_depthclearvalue =
r300_depth_clear_cb_value(surf->base.format, rgba);
width = surf->cbzb_width;
height = surf->cbzb_height;
r300->cbzb_clear = TRUE;
r300_mark_fb_state_dirty(r300, R300_CHANGED_CBZB_FLAG);
}
/* Clear. */
if (buffers) {
/* Clear using the blitter. */
r300_blitter_begin(r300, R300_CLEAR);
util_blitter_clear(r300->blitter,
width,
height,
fb->nr_cbufs,
buffers, rgba, depth, stencil);
r300_blitter_end(r300);
} else if (r300->zmask_clear.dirty) {
/* Just clear zmask and hiz now, this does not use a standard draw
* procedure. */
unsigned dwords;
/* Calculate zmask_clear and hiz_clear atom sizes. */
r300_update_hyperz_state(r300);
dwords = r300->zmask_clear.size +
(r300->hiz_clear.dirty ? r300->hiz_clear.size : 0) +
r300_get_num_cs_end_dwords(r300);
/* Reserve CS space. */
if (dwords > (R300_MAX_CMDBUF_DWORDS - r300->cs->cdw)) {
r300->context.flush(&r300->context, 0, NULL);
}
/* Emit clear packets. */
r300_emit_zmask_clear(r300, r300->zmask_clear.size,
r300->zmask_clear.state);
r300->zmask_clear.dirty = FALSE;
if (r300->hiz_clear.dirty) {
r300_emit_hiz_clear(r300, r300->hiz_clear.size,
r300->hiz_clear.state);
r300->hiz_clear.dirty = FALSE;
}
} else {
assert(0);
}
/* Disable CBZB clear. */
if (r300->cbzb_clear) {
r300->cbzb_clear = FALSE;
hyperz->zb_depthclearvalue = hyperz_dcv;
r300_mark_fb_state_dirty(r300, R300_CHANGED_CBZB_FLAG);
}
/* Enable fastfill and/or hiz.
*
* If we cleared zmask/hiz, it's in use now. The Hyper-Z state update
* looks if zmask/hiz is in use and enables fastfill accordingly. */
if (zstex &&
(zstex->zmask_in_use[fb->zsbuf->u.tex.level] ||
zstex->hiz_in_use[fb->zsbuf->u.tex.level])) {
r300_mark_atom_dirty(r300, &r300->hyperz_state);
}
}
