summaryrefslogtreecommitdiff
path: root/src/gallium/drivers/r300/r300_state.c
diff options
context:
space:
mode:
authorDave Airlie <airlied@redhat.com>2010-07-07 15:20:19 +0200
committerDave Airlie <airlied@redhat.com>2010-08-05 20:32:05 +1000
commit6eb2a7fbafd49e75b6cbbee57f23dda63eff73ef (patch)
treec52af432fc6562ece8a18592b67c9872897f4867 /src/gallium/drivers/r300/r300_state.c
parentccbd9ae7cc5b3fcda23fe79573e70b4fc40f3939 (diff)
r300g: implement hyper-z support. (v4)
This implements fast Z clear, Z compression, and HiZ support for r300->r500 GPUs. It also allows cbzb clears when fast Z clears are being used for the ZB. It requires a kernel with hyper-z support. Thanks to Marek Olšák <maraeo@gmail.com>, who started this off, and Alex Deucher at AMD for providing lots of hints. v2: squashed zmask ram size fix] squashed r300g/blitter: fix Z readback when compressed] v3: rebase around texture changes in master - .1 fix more bits v4: migrated to using u_mm in r300_texture to manage hiz/zmask rams consistently disabled HiZ when using OQ flush z-cache before turning hyper-z off update hyper-z state on dsa state change store depthclearvalue across cbzb clears and replace it afterwards. Signed-off-by: Dave Airlie <airlied@redhat.com>
Diffstat (limited to 'src/gallium/drivers/r300/r300_state.c')
-rw-r--r--src/gallium/drivers/r300/r300_state.c71
1 files changed, 56 insertions, 15 deletions
diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index a3383c3878..374aa254f0 100644
--- a/src/gallium/drivers/r300/r300_state.c
+++ b/src/gallium/drivers/r300/r300_state.c
@@ -25,6 +25,7 @@
#include "util/u_blitter.h"
#include "util/u_math.h"
+#include "util/u_mm.h"
#include "util/u_memory.h"
#include "util/u_pack_color.h"
@@ -43,6 +44,7 @@
#include "r300_texture.h"
#include "r300_vs.h"
#include "r300_winsys.h"
+#include "r300_hyperz.h"
/* r300_state: Functions used to intialize state context by translating
* Gallium state objects into semi-native r300 state objects. */
@@ -472,14 +474,14 @@ static void*
dsa->dsa = *state;
- /* Depth test setup. */
+ /* Depth test setup. - separate write mask depth for decomp flush */
+ if (state->depth.writemask) {
+ dsa->z_buffer_control |= R300_Z_WRITE_ENABLE;
+ }
+
if (state->depth.enabled) {
dsa->z_buffer_control |= R300_Z_ENABLE;
- if (state->depth.writemask) {
- dsa->z_buffer_control |= R300_Z_WRITE_ENABLE;
- }
-
dsa->z_stencil_control |=
(r300_translate_depth_stencil_function(state->depth.func) <<
R300_Z_FUNC_SHIFT);
@@ -592,6 +594,7 @@ static void r300_bind_dsa_state(struct pipe_context* pipe,
UPDATE_STATE(state, r300->dsa_state);
+ r300->hyperz_state.dirty = TRUE; /* Will be updated before the emission. */
r300_dsa_inject_stencilref(r300);
}
@@ -685,7 +688,8 @@ void r300_mark_fb_state_dirty(struct r300_context *r300,
/* What is marked as dirty depends on the enum r300_fb_state_change. */
r300->gpu_flush.dirty = TRUE;
r300->fb_state.dirty = TRUE;
- r300->hyperz_state.dirty = TRUE;
+ if (r300->rws->get_value(r300->rws, R300_CAN_HYPERZ))
+ r300->hyperz_state.dirty = TRUE;
if (change == R300_CHANGED_FB_STATE) {
r300->aa_state.dirty = TRUE;
@@ -698,7 +702,7 @@ void r300_mark_fb_state_dirty(struct r300_context *r300,
if (r300->cbzb_clear)
r300->fb_state.size += 10;
else if (state->zsbuf)
- r300->fb_state.size += r300->screen->caps.has_hiz ? 18 : 14;
+ r300->fb_state.size += r300->screen->caps.hiz_ram ? 18 : 14;
/* The size of the rest of atoms stays the same. */
}
@@ -710,8 +714,10 @@ static void
struct r300_context* r300 = r300_context(pipe);
struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state;
struct pipe_framebuffer_state *old_state = r300->fb_state.state;
+ boolean has_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
unsigned max_width, max_height, i;
uint32_t zbuffer_bpp = 0;
+ int blocksize;
if (r300->screen->caps.is_r500) {
max_width = max_height = 4096;
@@ -743,17 +749,52 @@ static void
r300_mark_fb_state_dirty(r300, R300_CHANGED_FB_STATE);
- /* Polygon offset depends on the zbuffer bit depth. */
+ r300->hiz_enable = false;
+ r300->z_fastfill = false;
+ r300->z_compression = false;
+
if (state->zsbuf) {
- switch (util_format_get_blocksize(state->zsbuf->texture->format)) {
- case 2:
- zbuffer_bpp = 16;
- break;
- case 4:
- zbuffer_bpp = 24;
- break;
+ blocksize = util_format_get_blocksize(state->zsbuf->texture->format);
+ switch (blocksize) {
+ case 2:
+ zbuffer_bpp = 16;
+ break;
+ case 4:
+ zbuffer_bpp = 24;
+ break;
}
+ if (has_hyperz) {
+ struct r300_surface *zs_surf = r300_surface(state->zsbuf);
+ struct r300_texture *tex;
+ int compress = r300->screen->caps.is_rv350 ? RV350_Z_COMPRESS_88 : R300_Z_COMPRESS_44;
+ int level = zs_surf->base.level;
+
+ tex = r300_texture(zs_surf->base.texture);
+
+ /* work out whether we can support hiz on this buffer */
+ r300_hiz_alloc_block(r300, zs_surf);
+
+ /* work out whether we can support zmask features on this buffer */
+ r300_zmask_alloc_block(r300, zs_surf, compress);
+
+ if (tex->hiz_mem[level]) {
+ r300->hiz_enable = 1;
+ }
+ if (tex->zmask_mem[level]) {
+ r300->z_fastfill = 1;
+ /* compression causes hangs on 16-bit */
+ if (zbuffer_bpp == 24)
+ r300->z_compression = compress;
+ }
+ DBG(r300, DBG_HYPERZ,
+ "hyper-z features: hiz: %d @ %08x z-compression: %d z-fastfill: %d @ %08x\n", r300->hiz_enable,
+ tex->hiz_mem[level] ? tex->hiz_mem[level]->ofs : 0xdeadbeef,
+ r300->z_compression, r300->z_fastfill,
+ tex->zmask_mem[level] ? tex->zmask_mem[level]->ofs : 0xdeadbeef);
+ }
+
+ /* Polygon offset depends on the zbuffer bit depth. */
if (r300->zbuffer_bpp != zbuffer_bpp) {
r300->zbuffer_bpp = zbuffer_bpp;