summaryrefslogtreecommitdiffstats
path: root/src/gallium/drivers/r300
diff options
context:
space:
mode:
Diffstat (limited to 'src/gallium/drivers/r300')
-rw-r--r--src/gallium/drivers/r300/r300_blit.c69
-rw-r--r--src/gallium/drivers/r300/r300_chipset.c54
-rw-r--r--src/gallium/drivers/r300/r300_chipset.h14
-rw-r--r--src/gallium/drivers/r300/r300_context.c36
-rw-r--r--src/gallium/drivers/r300/r300_context.h33
-rw-r--r--src/gallium/drivers/r300/r300_debug.c1
-rw-r--r--src/gallium/drivers/r300/r300_emit.c142
-rw-r--r--src/gallium/drivers/r300/r300_emit.h3
-rw-r--r--src/gallium/drivers/r300/r300_flush.c11
-rw-r--r--src/gallium/drivers/r300/r300_fs.c9
-rw-r--r--src/gallium/drivers/r300/r300_hyperz.c235
-rw-r--r--src/gallium/drivers/r300/r300_hyperz.h5
-rw-r--r--src/gallium/drivers/r300/r300_reg.h1
-rw-r--r--src/gallium/drivers/r300/r300_render.c162
-rw-r--r--src/gallium/drivers/r300/r300_render_stencilref.c51
-rw-r--r--src/gallium/drivers/r300/r300_screen.c3
-rw-r--r--src/gallium/drivers/r300/r300_screen.h1
-rw-r--r--src/gallium/drivers/r300/r300_state.c117
-rw-r--r--src/gallium/drivers/r300/r300_state_derived.c42
-rw-r--r--src/gallium/drivers/r300/r300_texture.c9
-rw-r--r--src/gallium/drivers/r300/r300_texture_desc.c22
-rw-r--r--src/gallium/drivers/r300/r300_winsys.h2
22 files changed, 796 insertions, 226 deletions
diff --git a/src/gallium/drivers/r300/r300_blit.c b/src/gallium/drivers/r300/r300_blit.c
index d125196b6dc..6f8d9abfc81 100644
--- a/src/gallium/drivers/r300/r300_blit.c
+++ b/src/gallium/drivers/r300/r300_blit.c
@@ -22,6 +22,7 @@
#include "r300_context.h"
#include "r300_texture.h"
+#include "r300_winsys.h"
#include "util/u_format.h"
#include "util/u_pack_color.h"
@@ -81,7 +82,7 @@ static void r300_blitter_end(struct r300_context *r300)
}
static uint32_t r300_depth_clear_cb_value(enum pipe_format format,
- const float* rgba)
+ const float* rgba)
{
union util_color uc;
util_pack_color(rgba, format, &uc);
@@ -98,6 +99,9 @@ static boolean r300_cbzb_clear_allowed(struct r300_context *r300,
struct pipe_framebuffer_state *fb =
(struct pipe_framebuffer_state*)r300->fb_state.state;
+ if (r300->z_fastfill)
+ clear_buffers &= ~(PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL);
+
/* Only color clear allowed, and only one colorbuffer. */
if (clear_buffers != PIPE_CLEAR_COLOR || fb->nr_cbufs != 1)
return FALSE;
@@ -105,6 +109,23 @@ static boolean r300_cbzb_clear_allowed(struct r300_context *r300,
return r300_surface(fb->cbufs[0])->cbzb_allowed;
}
+static uint32_t r300_depth_clear_value(enum pipe_format format,
+ double depth, unsigned stencil)
+{
+ switch (format) {
+ case PIPE_FORMAT_Z16_UNORM:
+ case PIPE_FORMAT_X8Z24_UNORM:
+ return util_pack_z(format, depth);
+
+ case PIPE_FORMAT_S8_USCALED_Z24_UNORM:
+ return util_pack_z_stencil(format, depth, stencil);
+
+ default:
+ assert(0);
+ return 0;
+ }
+}
+
/* Clear currently bound buffers. */
static void r300_clear(struct pipe_context* pipe,
unsigned buffers,
@@ -154,6 +175,22 @@ static void r300_clear(struct pipe_context* pipe,
(struct r300_hyperz_state*)r300->hyperz_state.state;
uint32_t width = fb->width;
uint32_t height = fb->height;
+ boolean has_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
+ uint32_t hyperz_dcv = 0;
+
+ /* Enable fast Z clear.
+ * The zbuffer must be in micro-tiled mode, otherwise it locks up. */
+ if ((buffers & (PIPE_CLEAR_DEPTH|PIPE_CLEAR_STENCIL)) && has_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 (r300->z_compression || r300->z_fastfill)
+ r300->zmask_clear.dirty = TRUE;
+ if (r300->hiz_enable)
+ r300->hiz_clear.dirty = TRUE;
+ }
/* Enable CBZB clear. */
if (r300_cbzb_clear_allowed(r300, buffers)) {
@@ -181,6 +218,7 @@ static void r300_clear(struct pipe_context* pipe,
/* 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);
}
@@ -221,6 +259,29 @@ static void r300_clear_depth_stencil(struct pipe_context *pipe,
r300_blitter_end(r300);
}
+/* Clear a region of a depth stencil surface. */
+static void r300_flush_depth_stencil(struct pipe_context *pipe,
+ struct pipe_resource *dst,
+ struct pipe_subresource subdst)
+{
+ struct r300_context *r300 = r300_context(pipe);
+ struct pipe_surface *dstsurf;
+ struct r300_texture *tex = r300_texture(dst);
+
+ /* only flush the zmask if we have one attached to this texture */
+ if (!tex->zmask_mem[subdst.level])
+ return;
+
+ dstsurf = pipe->screen->get_tex_surface(pipe->screen, dst,
+ subdst.face, subdst.level, 0,
+ PIPE_BIND_DEPTH_STENCIL);
+ 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;
+}
+
/* 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,
@@ -252,7 +313,7 @@ static void r300_resource_copy_region(struct pipe_context *pipe,
{
enum pipe_format old_format = dst->format;
enum pipe_format new_format = old_format;
-
+ boolean is_depth;
if (!pipe->screen->is_format_supported(pipe->screen,
old_format, src->target,
src->nr_samples,
@@ -279,6 +340,10 @@ static void r300_resource_copy_region(struct pipe_context *pipe,
}
}
+ is_depth = util_format_get_component_bits(src->format, UTIL_FORMAT_COLORSPACE_ZS, 0) != 0;
+ if (is_depth) {
+ r300_flush_depth_stencil(pipe, src, subsrc);
+ }
if (old_format != new_format) {
dst->format = new_format;
src->format = new_format;
diff --git a/src/gallium/drivers/r300/r300_chipset.c b/src/gallium/drivers/r300/r300_chipset.c
index 21f3b9d2610..48c24092114 100644
--- a/src/gallium/drivers/r300/r300_chipset.c
+++ b/src/gallium/drivers/r300/r300_chipset.c
@@ -36,7 +36,7 @@ void r300_parse_chipset(struct r300_capabilities* caps)
caps->num_vert_fpus = 2;
caps->num_tex_units = 16;
caps->has_tcl = debug_get_bool_option("RADEON_NO_TCL", FALSE) ? FALSE : TRUE;
- caps->has_hiz = TRUE;
+ caps->hiz_ram = 0;
caps->is_r400 = FALSE;
caps->is_r500 = FALSE;
caps->high_second_pipe = FALSE;
@@ -49,6 +49,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
caps->family = CHIP_FAMILY_R300;
caps->high_second_pipe = TRUE;
caps->num_vert_fpus = 4;
+ caps->hiz_ram = R300_HIZ_LIMIT;
+ caps->zmask_ram = PIPE_ZMASK_SIZE;
break;
case 0x4145:
@@ -61,6 +63,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
caps->family = CHIP_FAMILY_R300;
caps->high_second_pipe = TRUE;
caps->num_vert_fpus = 4;
+ caps->hiz_ram = R300_HIZ_LIMIT;
+ caps->zmask_ram = PIPE_ZMASK_SIZE;
break;
case 0x4150:
@@ -77,8 +81,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
case 0x4E54:
case 0x4E56:
caps->family = CHIP_FAMILY_RV350;
- caps->has_hiz = FALSE;
caps->high_second_pipe = TRUE;
+ caps->zmask_ram = RV3xx_ZMASK_SIZE;
break;
case 0x4148:
@@ -91,12 +95,16 @@ void r300_parse_chipset(struct r300_capabilities* caps)
caps->family = CHIP_FAMILY_R350;
caps->high_second_pipe = TRUE;
caps->num_vert_fpus = 4;
+ caps->hiz_ram = R300_HIZ_LIMIT;
+ caps->zmask_ram = PIPE_ZMASK_SIZE;
break;
case 0x4E4A:
caps->family = CHIP_FAMILY_R360;
caps->high_second_pipe = TRUE;
caps->num_vert_fpus = 4;
+ caps->hiz_ram = R300_HIZ_LIMIT;
+ caps->zmask_ram = PIPE_ZMASK_SIZE;
break;
case 0x5460:
@@ -108,8 +116,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
case 0x5B64:
case 0x5B65:
caps->family = CHIP_FAMILY_RV370;
- caps->has_hiz = FALSE;
caps->high_second_pipe = TRUE;
+ caps->zmask_ram = RV3xx_ZMASK_SIZE;
break;
case 0x3150:
@@ -120,6 +128,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
case 0x3E54:
caps->family = CHIP_FAMILY_RV380;
caps->high_second_pipe = TRUE;
+ caps->hiz_ram = R300_HIZ_LIMIT;
+ caps->zmask_ram = RV3xx_ZMASK_SIZE;
break;
case 0x4A48:
@@ -135,6 +145,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
caps->family = CHIP_FAMILY_R420;
caps->num_vert_fpus = 6;
caps->is_r400 = TRUE;
+ caps->hiz_ram = R300_HIZ_LIMIT;
+ caps->zmask_ram = PIPE_ZMASK_SIZE;
break;
case 0x5548:
@@ -149,6 +161,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
caps->family = CHIP_FAMILY_R423;
caps->num_vert_fpus = 6;
caps->is_r400 = TRUE;
+ caps->hiz_ram = R300_HIZ_LIMIT;
+ caps->zmask_ram = PIPE_ZMASK_SIZE;
break;
case 0x554C:
@@ -161,6 +175,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
caps->family = CHIP_FAMILY_R430;
caps->num_vert_fpus = 6;
caps->is_r400 = TRUE;
+ caps->hiz_ram = R300_HIZ_LIMIT;
+ caps->zmask_ram = PIPE_ZMASK_SIZE;
break;
case 0x5D4C:
@@ -172,6 +188,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
caps->family = CHIP_FAMILY_R480;
caps->num_vert_fpus = 6;
caps->is_r400 = TRUE;
+ caps->hiz_ram = R300_HIZ_LIMIT;
+ caps->zmask_ram = PIPE_ZMASK_SIZE;
break;
case 0x4B48:
@@ -182,6 +200,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
caps->family = CHIP_FAMILY_R481;
caps->num_vert_fpus = 6;
caps->is_r400 = TRUE;
+ caps->hiz_ram = R300_HIZ_LIMIT;
+ caps->zmask_ram = PIPE_ZMASK_SIZE;
break;
case 0x5E4C:
@@ -199,34 +219,36 @@ void r300_parse_chipset(struct r300_capabilities* caps)
caps->family = CHIP_FAMILY_RV410;
caps->num_vert_fpus = 6;
caps->is_r400 = TRUE;
+ caps->hiz_ram = R300_HIZ_LIMIT;
+ caps->zmask_ram = PIPE_ZMASK_SIZE;
break;
case 0x5954:
case 0x5955:
caps->family = CHIP_FAMILY_RS480;
- caps->has_hiz = FALSE;
caps->has_tcl = FALSE;
+ caps->zmask_ram = RV3xx_ZMASK_SIZE;
break;
case 0x5974:
case 0x5975:
caps->family = CHIP_FAMILY_RS482;
- caps->has_hiz = FALSE;
caps->has_tcl = FALSE;
+ caps->zmask_ram = RV3xx_ZMASK_SIZE;
break;
case 0x5A41:
case 0x5A42:
caps->family = CHIP_FAMILY_RS400;
- caps->has_hiz = FALSE;
caps->has_tcl = FALSE;
+ caps->zmask_ram = RV3xx_ZMASK_SIZE;
break;
case 0x5A61:
case 0x5A62:
caps->family = CHIP_FAMILY_RC410;
- caps->has_hiz = FALSE;
caps->has_tcl = FALSE;
+ caps->zmask_ram = RV3xx_ZMASK_SIZE;
break;
case 0x791E:
@@ -234,6 +256,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
caps->family = CHIP_FAMILY_RS690;
caps->has_tcl = FALSE;
caps->is_r400 = TRUE;
+ caps->hiz_ram = R300_HIZ_LIMIT;
+ caps->zmask_ram = PIPE_ZMASK_SIZE;
break;
case 0x793F:
@@ -242,6 +266,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
caps->family = CHIP_FAMILY_RS600;
caps->has_tcl = FALSE;
caps->is_r400 = TRUE;
+ caps->hiz_ram = R300_HIZ_LIMIT;
+ caps->zmask_ram = PIPE_ZMASK_SIZE;
break;
case 0x796C:
@@ -251,6 +277,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
caps->family = CHIP_FAMILY_RS740;
caps->has_tcl = FALSE;
caps->is_r400 = TRUE;
+ caps->hiz_ram = R300_HIZ_LIMIT;
+ caps->zmask_ram = PIPE_ZMASK_SIZE;
break;
case 0x7100:
@@ -270,6 +298,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
caps->family = CHIP_FAMILY_R520;
caps->num_vert_fpus = 8;
caps->is_r500 = TRUE;
+ caps->hiz_ram = R300_HIZ_LIMIT;
+ caps->zmask_ram = PIPE_ZMASK_SIZE;
break;
case 0x7140:
@@ -313,6 +343,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
caps->family = CHIP_FAMILY_RV515;
caps->num_vert_fpus = 2;
caps->is_r500 = TRUE;
+ caps->hiz_ram = R300_HIZ_LIMIT;
+ caps->zmask_ram = PIPE_ZMASK_SIZE;
break;
case 0x71C0:
@@ -334,6 +366,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
caps->family = CHIP_FAMILY_RV530;
caps->num_vert_fpus = 5;
caps->is_r500 = TRUE;
+ /*caps->hiz_ram = RV530_HIZ_LIMIT;*/
+ caps->zmask_ram = PIPE_ZMASK_SIZE;
break;
case 0x7240:
@@ -354,12 +388,16 @@ void r300_parse_chipset(struct r300_capabilities* caps)
caps->family = CHIP_FAMILY_R580;
caps->num_vert_fpus = 8;
caps->is_r500 = TRUE;
+ caps->hiz_ram = RV530_HIZ_LIMIT;
+ caps->zmask_ram = PIPE_ZMASK_SIZE;
break;
case 0x7280:
caps->family = CHIP_FAMILY_RV570;
caps->num_vert_fpus = 8;
caps->is_r500 = TRUE;
+ caps->hiz_ram = RV530_HIZ_LIMIT;
+ caps->zmask_ram = PIPE_ZMASK_SIZE;
break;
case 0x7281:
@@ -376,6 +414,8 @@ void r300_parse_chipset(struct r300_capabilities* caps)
caps->family = CHIP_FAMILY_RV560;
caps->num_vert_fpus = 8;
caps->is_r500 = TRUE;
+ caps->hiz_ram = RV530_HIZ_LIMIT;
+ caps->zmask_ram = PIPE_ZMASK_SIZE;
break;
default:
diff --git a/src/gallium/drivers/r300/r300_chipset.h b/src/gallium/drivers/r300/r300_chipset.h
index 65750f54e71..e7ca642b4f3 100644
--- a/src/gallium/drivers/r300/r300_chipset.h
+++ b/src/gallium/drivers/r300/r300_chipset.h
@@ -25,6 +25,14 @@
#include "pipe/p_compiler.h"
+/* these are sizes in dwords */
+#define R300_HIZ_LIMIT 10240
+#define RV530_HIZ_LIMIT 15360
+
+/* rv3xx have only one pipe */
+#define PIPE_ZMASK_SIZE 4096
+#define RV3xx_ZMASK_SIZE 5120
+
/* Structure containing all the possible information about a specific Radeon
* in the R3xx, R4xx, and R5xx families. */
struct r300_capabilities {
@@ -42,8 +50,10 @@ struct r300_capabilities {
unsigned num_tex_units;
/* Whether or not TCL is physically present */
boolean has_tcl;
- /* Some chipsets do not have HiZ RAM. */
- boolean has_hiz;
+ /* Some chipsets do not have HiZ RAM - other have varying amounts . */
+ int hiz_ram;
+ /* some chipsets have zmask ram per pipe some don't */
+ int zmask_ram;
/* Whether or not this is RV350 or newer, including all r400 and r500
* chipsets. The differences compared to the oldest r300 chips are:
* - Blend LTE/GTE thresholds
diff --git a/src/gallium/drivers/r300/r300_context.c b/src/gallium/drivers/r300/r300_context.c
index df903590583..e8b6c4f7af8 100644
--- a/src/gallium/drivers/r300/r300_context.c
+++ b/src/gallium/drivers/r300/r300_context.c
@@ -30,6 +30,7 @@
#include "r300_cb.h"
#include "r300_context.h"
#include "r300_emit.h"
+#include "r300_hyperz.h"
#include "r300_screen.h"
#include "r300_screen_buffer.h"
#include "r300_winsys.h"
@@ -114,6 +115,10 @@ static void r300_destroy_context(struct pipe_context* context)
u_upload_destroy(r300->upload_vb);
u_upload_destroy(r300->upload_ib);
+ /* setup hyper-z mm */
+ if (r300->rws->get_value(r300->rws, R300_CAN_HYPERZ))
+ r300_hyperz_destroy_mm(r300);
+
translate_cache_destroy(r300->tran.translate_cache);
r300_release_referenced_objects(r300);
@@ -166,6 +171,9 @@ static void r300_setup_atoms(struct r300_context* r300)
boolean is_r500 = r300->screen->caps.is_r500;
boolean has_tcl = r300->screen->caps.has_tcl;
boolean drm_2_3_0 = r300->rws->get_value(r300->rws, R300_VID_DRM_2_3_0);
+ boolean drm_2_6_0 = r300->rws->get_value(r300->rws, R300_VID_DRM_2_6_0);
+ boolean has_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
+ boolean has_hiz_ram = r300->screen->caps.hiz_ram > 0;
/* Create the actual atom list.
*
@@ -188,8 +196,8 @@ static void r300_setup_atoms(struct r300_context* r300)
R300_INIT_ATOM(gpu_flush, 9);
R300_INIT_ATOM(aa_state, 4);
R300_INIT_ATOM(fb_state, 0);
+ R300_INIT_ATOM(hyperz_state, is_r500 || (is_rv350 && drm_2_6_0) ? 10 : 8);
/* ZB (unpipelined), SC. */
- R300_INIT_ATOM(hyperz_state, 6);
R300_INIT_ATOM(ztop_state, 2);
/* ZB, FG. */
R300_INIT_ATOM(dsa_state, is_r500 ? 8 : 6);
@@ -220,6 +228,13 @@ static void r300_setup_atoms(struct r300_context* r300)
/* TX. */
R300_INIT_ATOM(texture_cache_inval, 2);
R300_INIT_ATOM(textures_state, 0);
+ if (has_hyperz) {
+ /* HiZ Clear */
+ if (has_hiz_ram)
+ R300_INIT_ATOM(hiz_clear, 0);
+ /* zmask clear */
+ R300_INIT_ATOM(zmask_clear, 0);
+ }
/* ZB (unpipelined), SU. */
R300_INIT_ATOM(query_start, 4);
@@ -282,8 +297,7 @@ static void r300_init_states(struct pipe_context *pipe)
(struct r300_vap_invariant_state*)r300->vap_invariant_state.state;
struct r300_invariant_state *invariant =
(struct r300_invariant_state*)r300->invariant_state.state;
- struct r300_hyperz_state *hyperz =
- (struct r300_hyperz_state*)r300->hyperz_state.state;
+
CB_LOCALS;
pipe->set_blend_color(pipe, &bc);
@@ -351,10 +365,20 @@ static void r300_init_states(struct pipe_context *pipe)
/* Initialize the hyperz state. */
{
- BEGIN_CB(&hyperz->cb_begin, r300->hyperz_state.size);
+ struct r300_hyperz_state *hyperz =
+ (struct r300_hyperz_state*)r300->hyperz_state.state;
+ BEGIN_CB(&hyperz->cb_flush_begin, r300->hyperz_state.size);
+ OUT_CB_REG(R300_ZB_ZCACHE_CTLSTAT,
+ R300_ZB_ZCACHE_CTLSTAT_ZC_FLUSH_FLUSH_AND_FREE);
OUT_CB_REG(R300_ZB_BW_CNTL, 0);
OUT_CB_REG(R300_ZB_DEPTHCLEARVALUE, 0);
OUT_CB_REG(R300_SC_HYPERZ, R300_SC_HYPERZ_ADJ_2);
+
+ if (r300->screen->caps.is_r500 ||
+ (r300->screen->caps.is_rv350 &&
+ r300->rws->get_value(r300->rws, R300_VID_DRM_2_6_0))) {
+ OUT_CB_REG(R300_GB_Z_PEQ_CONFIG, 0);
+ }
END_CB;
}
}
@@ -415,6 +439,10 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
rws->cs_set_flush(r300->cs, r300_flush_cb, r300);
+ /* setup hyper-z mm */
+ if (r300->rws->get_value(r300->rws, R300_CAN_HYPERZ))
+ r300_hyperz_init_mm(r300);
+
r300->upload_ib = u_upload_create(&r300->context,
32 * 1024, 16,
PIPE_BIND_INDEX_BUFFER);
diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h
index b4256c62786..d86a5c8fc98 100644
--- a/src/gallium/drivers/r300/r300_context.h
+++ b/src/gallium/drivers/r300/r300_context.h
@@ -106,13 +106,19 @@ struct r300_dsa_state {
};
struct r300_hyperz_state {
+ int current_func; /* -1 after a clear before first op */
+ int flush;
/* This is actually a command buffer with named dwords. */
+ uint32_t cb_flush_begin;
+ uint32_t zb_zcache_ctlstat; /* R300_ZB_CACHE_CNTL */
uint32_t cb_begin;
uint32_t zb_bw_cntl; /* R300_ZB_BW_CNTL */
uint32_t cb_reg1;
uint32_t zb_depthclearvalue; /* R300_ZB_DEPTHCLEARVALUE */
uint32_t cb_reg2;
uint32_t sc_hyperz; /* R300_SC_HYPERZ */
+ uint32_t cb_reg3;
+ uint32_t gb_z_peq_config; /* R300_GB_Z_PEQ_CONFIG: 0x4028 */
};
struct r300_gpu_flush {
@@ -321,6 +327,7 @@ struct r300_surface {
/* Whether the CBZB clear is allowed on the surface. */
boolean cbzb_allowed;
+
};
struct r300_texture_desc {
@@ -387,6 +394,10 @@ struct r300_texture {
/* All bits should be filled in. */
struct r300_texture_fb_state fb_state;
+ /* hyper-z memory allocs */
+ struct mem_block *hiz_mem[R300_MAX_TEXTURE_LEVELS];
+ struct mem_block *zmask_mem[R300_MAX_TEXTURE_LEVELS];
+
/* This is the level tiling flags were last time set for.
* It's used to prevent redundant tiling-flags changes from happening.*/
unsigned surface_level;
@@ -512,6 +523,10 @@ struct r300_context {
struct r300_atom texture_cache_inval;
/* GPU flush. */
struct r300_atom gpu_flush;
+ /* HiZ clear */
+ struct r300_atom hiz_clear;
+ /* zmask clear */
+ struct r300_atom zmask_clear;
/* Invariant state. This must be emitted to get the engine started. */
struct r300_atom invariant_state;
@@ -524,6 +539,8 @@ struct r300_context {
struct r300_vertex_element_state *velems;
bool any_user_vbs;
+ struct pipe_index_buffer index_buffer;
+
/* Vertex info for Draw. */
struct vertex_info vertex_info;
@@ -547,8 +564,19 @@ struct r300_context {
boolean two_sided_color;
/* Incompatible vertex buffer layout? (misaligned stride or buffer_offset) */
boolean incompatible_vb_layout;
-
+ /* Whether fast zclear is enabled. */
+ boolean z_fastfill;
+#define R300_Z_COMPRESS_44 1
+#define RV350_Z_COMPRESS_88 2
+ int z_compression;
+ boolean hiz_enable;
boolean cbzb_clear;
+ boolean z_decomp_rd;
+
+ /* two mem block managers for hiz/zmask ram space */
+ struct mem_block *hiz_mm;
+ struct mem_block *zmask_mm;
+
/* upload managers */
struct u_upload_mgr *upload_vb;
struct u_upload_mgr *upload_ib;
@@ -619,7 +647,8 @@ void r300_plug_in_stencil_ref_fallback(struct r300_context *r300);
/* r300_state.c */
enum r300_fb_state_change {
R300_CHANGED_FB_STATE = 0,
- R300_CHANGED_CBZB_FLAG
+ R300_CHANGED_CBZB_FLAG,
+ R300_CHANGED_ZCLEAR_FLAG
};
void r300_mark_fb_state_dirty(struct r300_context *r300,
diff --git a/src/gallium/drivers/r300/r300_debug.c b/src/gallium/drivers/r300/r300_debug.c
index 053a64ea6d7..c3e157e99af 100644
--- a/src/gallium/drivers/r300/r300_debug.c
+++ b/src/gallium/drivers/r300/r300_debug.c
@@ -44,6 +44,7 @@ static const struct debug_named_value debug_options[] = {
{ "notiling", DBG_NO_TILING, "Disable tiling (for benchmarking)" },
{ "noimmd", DBG_NO_IMMD, "Disable immediate mode (for benchmarking)" },
{ "stats", DBG_STATS, "Gather statistics" },
+ { "hyperz", DBG_HYPERZ, "HyperZ (for debugging)" },
/* must be last */
DEBUG_NAMED_VALUE_END
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index 36a26a78717..17e180a79ac 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -25,6 +25,7 @@
#include "util/u_format.h"
#include "util/u_math.h"
+#include "util/u_mm.h"
#include "util/u_simple_list.h"
#include "r300_context.h"
@@ -329,6 +330,7 @@ void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state)
struct pipe_framebuffer_state* fb = (struct pipe_framebuffer_state*)state;
struct r300_surface* surf;
unsigned i;
+ boolean has_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
CS_LOCALS(r300);
BEGIN_CS(size);
@@ -364,6 +366,10 @@ void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state)
OUT_CS_REG_SEQ(R300_ZB_DEPTHPITCH, 1);
OUT_CS_RELOC(surf->buffer, surf->cbzb_pitch, 0, surf->domain);
+
+ DBG(r300, DBG_CBZB,
+ "CBZB clearing cbuf %08x %08x\n", surf->cbzb_format,
+ surf->cbzb_pitch);
}
/* Set up a zbuffer. */
else if (fb->zsbuf) {
@@ -377,15 +383,32 @@ void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state)
OUT_CS_REG_SEQ(R300_ZB_DEPTHPITCH, 1);
OUT_CS_RELOC(surf->buffer, surf->pitch, 0, surf->domain);
- /* HiZ RAM. */
- if (r300->screen->caps.has_hiz) {
- OUT_CS_REG(R300_ZB_HIZ_OFFSET, 0);
- OUT_CS_REG(R300_ZB_HIZ_PITCH, 0);
+ if (has_hyperz) {
+ uint32_t surf_pitch;
+ struct r300_texture *tex;
+ int level = surf->base.level;
+ tex = r300_texture(surf->base.texture);
+
+ surf_pitch = surf->pitch & R300_DEPTHPITCH_MASK;
+ /* HiZ RAM. */
+ if (r300->screen->caps.hiz_ram) {
+ if (tex->hiz_mem[level]) {
+ OUT_CS_REG(R300_ZB_HIZ_OFFSET, tex->hiz_mem[level]->ofs);
+ OUT_CS_REG(R300_ZB_HIZ_PITCH, surf_pitch);
+ } else {
+ OUT_CS_REG(R300_ZB_HIZ_OFFSET, 0);
+ OUT_CS_REG(R300_ZB_HIZ_PITCH, 0);
+ }
+ }
+ /* Z Mask RAM. (compressed zbuffer) */
+ if (tex->zmask_mem[level]) {
+ OUT_CS_REG(R300_ZB_ZMASK_OFFSET, tex->zmask_mem[level]->ofs);
+ OUT_CS_REG(R300_ZB_ZMASK_PITCH, surf_pitch);
+ } else {
+ OUT_CS_REG(R300_ZB_ZMASK_OFFSET, 0);
+ OUT_CS_REG(R300_ZB_ZMASK_PITCH, 0);
+ }
}
-
- /* Z Mask RAM. (compressed zbuffer) */
- OUT_CS_REG(R300_ZB_ZMASK_OFFSET, 0);
- OUT_CS_REG(R300_ZB_ZMASK_PITCH, 0);
}
END_CS;
@@ -394,8 +417,12 @@ void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state)
void r300_emit_hyperz_state(struct r300_context *r300,
unsigned size, void *state)
{
+ struct r300_hyperz_state *z = state;
CS_LOCALS(r300);
- WRITE_CS_TABLE(state, size);
+ if (z->flush)
+ WRITE_CS_TABLE(&z->cb_flush_begin, size);
+ else
+ WRITE_CS_TABLE(&z->cb_begin, size - 2);
}
void r300_emit_hyperz_end(struct r300_context *r300)
@@ -403,9 +430,11 @@ void r300_emit_hyperz_end(struct r300_context *r300)
struct r300_hyperz_state z =
*(struct r300_hyperz_state*)r300->hyperz_state.state;
+ z.flush = 1;
z.zb_bw_cntl = 0;
z.zb_depthclearvalue = 0;
z.sc_hyperz = R300_SC_HYPERZ_ADJ_2;
+ z.gb_z_peq_config = 0;
r300_emit_hyperz_state(r300, r300->hyperz_state.size, &z);
}
@@ -943,6 +972,101 @@ void r300_emit_viewport_state(struct r300_context* r300,
END_CS;
}
+static void r300_emit_hiz_line_clear(struct r300_context *r300, int start, uint16_t count, uint32_t val)
+{
+ CS_LOCALS(r300);
+ BEGIN_CS(4);
+ OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_HIZ, 2);
+ OUT_CS(start);
+ OUT_CS(count);
+ OUT_CS(val);
+ END_CS;
+}
+
+static void r300_emit_zmask_line_clear(struct r300_context *r300, int start, uint16_t count, uint32_t val)
+{
+ CS_LOCALS(r300);
+ BEGIN_CS(4);
+ OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_ZMASK, 2);
+ OUT_CS(start);
+ OUT_CS(count);
+ OUT_CS(val);
+ END_CS;
+}
+
+#define ALIGN_DIVUP(x, y) (((x) + (y) - 1) / (y))
+
+void r300_emit_hiz_clear(struct r300_context *r300, unsigned size, void *state)
+{
+ struct pipe_framebuffer_state *fb =
+ (struct pipe_framebuffer_state*)r300->fb_state.state;
+ struct r300_hyperz_state *z =
+ (struct r300_hyperz_state*)r300->hyperz_state.state;
+ struct r300_screen* r300screen = r300->screen;
+ uint32_t stride, offset = 0, height, offset_shift;
+ struct r300_texture* tex;
+ int i;
+
+ tex = r300_texture(fb->zsbuf->texture);
+ stride = tex->desc.stride_in_pixels[fb->zsbuf->level];
+
+ /* convert from pixels to 4x4 blocks */
+ stride = ALIGN_DIVUP(stride, 4);
+
+ stride = ALIGN_DIVUP(stride, r300screen->caps.num_frag_pipes);
+ /* there are 4 blocks per dwords */
+ stride = ALIGN_DIVUP(stride, 4);
+
+ height = ALIGN_DIVUP(fb->zsbuf->height, 4);
+
+ offset_shift = 2;
+ offset_shift += (r300screen->caps.num_frag_pipes / 2);
+
+ for (i = 0; i < height; i++) {
+ offset = i * stride;
+ offset <<= offset_shift;
+ r300_emit_hiz_line_clear(r300, offset, stride, 0xffffffff);
+ }
+ z->current_func = -1;
+}
+
+void r300_emit_zmask_clear(struct r300_context *r300, unsigned size, void *state)
+{
+ struct pipe_framebuffer_state *fb =
+ (struct pipe_framebuffer_state*)r300->fb_state.state;
+ struct r300_screen* r300screen = r300->screen;
+ uint32_t stride, offset = 0;
+ struct r300_texture* tex;
+ uint32_t i, height;
+ int mult, offset_shift;
+
+ tex = r300_texture(fb->zsbuf->texture);
+ stride = tex->desc.stride_in_pixels[fb->zsbuf->level];
+
+ if (r300->z_compression == RV350_Z_COMPRESS_88)
+ mult = 8;
+ else
+ mult = 4;
+
+ height = ALIGN_DIVUP(fb->zsbuf->height, mult);
+
+ offset_shift = 4;
+ offset_shift += (r300screen->caps.num_frag_pipes / 2);
+ stride = ALIGN_DIVUP(stride, r300screen->caps.num_frag_pipes);
+
+ /* okay have width in pixels - divide by block width */
+ stride = ALIGN_DIVUP(stride, mult);
+ /* have width in blocks - divide by number of fragment pipes screen width */
+ /* 16 blocks per dword */
+ stride = ALIGN_DIVUP(stride, 16);
+
+ for (i = 0; i < height; i++) {
+ offset = i * stride;
+ offset <<= offset_shift;
+ r300_emit_zmask_line_clear(r300, offset, stride, 0x0);//0xffffffff);
+ }
+}
+
void r300_emit_ztop_state(struct r300_context* r300,
unsigned size, void* state)
{
diff --git a/src/gallium/drivers/r300/r300_emit.h b/src/gallium/drivers/r300/r300_emit.h
index 5d05039669f..2f2c2f2dcb4 100644
--- a/src/gallium/drivers/r300/r300_emit.h
+++ b/src/gallium/drivers/r300/r300_emit.h
@@ -112,6 +112,9 @@ void r300_emit_texture_cache_inval(struct r300_context* r300, unsigned size, voi
void r300_emit_invariant_state(struct r300_context *r300,
unsigned size, void *state);
+void r300_emit_hiz_clear(struct r300_context *r300, unsigned size, void *state);
+void r300_emit_zmask_clear(struct r300_context *r300, unsigned size, void *state);
+
unsigned r300_get_num_dirty_dwords(struct r300_context *r300);
/* Emit all dirty state. */
diff --git a/src/gallium/drivers/r300/r300_flush.c b/src/gallium/drivers/r300/r300_flush.c
index ae7b5759e78..7fed9b5d074 100644
--- a/src/gallium/drivers/r300/r300_flush.c
+++ b/src/gallium/drivers/r300/r300_flush.c
@@ -43,16 +43,9 @@ static void r300_flush(struct pipe_context* pipe,
u_upload_flush(r300->upload_vb);
u_upload_flush(r300->upload_ib);
- /* We probably need to flush Draw, but we may have been called from
- * within Draw. This feels kludgy, but it might be the best thing.
- *
- * Of course, the best thing is to kill Draw with fire. :3 */
- if (r300->draw && !r300->draw->flushing) {
- draw_flush(r300->draw);
- }
-
if (r300->dirty_hw) {
- r300_emit_hyperz_end(r300);
+ if (r300->rws->get_value(r300->rws, R300_CAN_HYPERZ))
+ r300_emit_hyperz_end(r300);
r300_emit_query_end(r300);
r300->flush_counter++;
diff --git a/src/gallium/drivers/r300/r300_fs.c b/src/gallium/drivers/r300/r300_fs.c
index db5269912e2..87ff49a90c7 100644
--- a/src/gallium/drivers/r300/r300_fs.c
+++ b/src/gallium/drivers/r300/r300_fs.c
@@ -248,13 +248,18 @@ static void r300_emit_fs_code_to_buffer(
shader->cb_code_size = 19 +
((code->inst_end + 1) * 6) +
- imm_count * 7;
+ imm_count * 7 +
+ code->int_constant_count * 2;
NEW_CB(shader->cb_code, shader->cb_code_size);
OUT_CB_REG(R500_US_CONFIG, R500_ZERO_TIMES_ANYTHING_EQUALS_ZERO);
OUT_CB_REG(R500_US_PIXSIZE, code->max_temp_idx);
OUT_CB_REG(R500_US_FC_CTRL, code->us_fc_ctrl);
- OUT_CB_REG(R500_US_CODE_RANGE,
+ for(i = 0; i < code->int_constant_count; i++){
+ OUT_CB_REG(R500_US_FC_INT_CONST_0 + (i * 4),
+ code->int_constants[i]);
+ }
+ OUT_CB_REG(R500_US_CODE_RANGE,
R500_US_CODE_RANGE_ADDR(0) | R500_US_CODE_RANGE_SIZE(code->inst_end));
OUT_CB_REG(R500_US_CODE_OFFSET, 0);
OUT_CB_REG(R500_US_CODE_ADDR,
diff --git a/src/gallium/drivers/r300/r300_hyperz.c b/src/gallium/drivers/r300/r300_hyperz.c
index e9528956019..10e440ce306 100644
--- a/src/gallium/drivers/r300/r300_hyperz.c
+++ b/src/gallium/drivers/r300/r300_hyperz.c
@@ -21,25 +21,156 @@
* OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
* USE OR OTHER DEALINGS IN THE SOFTWARE. */
+#include "util/u_format.h"
+#include "util/u_mm.h"
#include "r300_context.h"
#include "r300_hyperz.h"
#include "r300_reg.h"
#include "r300_fs.h"
+/*
+ HiZ rules - taken from various docs
+ 1. HiZ only works on depth values
+ 2. Cannot HiZ if stencil fail or zfail is !KEEP
+ 3. on R300/400, HiZ is disabled if depth test is EQUAL
+ 4. comparison changes without clears usually mean disabling HiZ
+*/
/*****************************************************************************/
/* The HyperZ setup */
/*****************************************************************************/
+static bool r300_get_sc_hz_max(struct r300_context *r300)
+{
+ struct r300_dsa_state *dsa_state = r300->dsa_state.state;
+ int func = dsa_state->z_stencil_control & 0x7;
+ int ret = R300_SC_HYPERZ_MIN;
+
+ if (func >= 4 && func <= 7)
+ ret = R300_SC_HYPERZ_MAX;
+ return ret;
+}
+
+static bool r300_zfunc_same_direction(int func1, int func2)
+{
+ /* func1 is less/lessthan */
+ if (func1 == 1 || func1 == 2)
+ if (func2 == 3 || func2 == 4 || func2 == 5)
+ return FALSE;
+
+ if (func2 == 1 || func2 == 2)
+ if (func1 == 4 || func1 == 5)
+ return FALSE;
+ return TRUE;
+}
+
+static int r300_get_hiz_min(struct r300_context *r300)
+{
+ struct r300_dsa_state *dsa_state = r300->dsa_state.state;
+ int func = dsa_state->z_stencil_control & 0x7;
+ int ret = R300_HIZ_MIN;
+
+ if (func == 1 || func == 2)
+ ret = R300_HIZ_MAX;
+ return ret;
+}
+
+static boolean r300_dsa_stencil_op_not_keep(struct pipe_stencil_state *s)
+{
+ if (s->enabled && (s->fail_op != PIPE_STENCIL_OP_KEEP ||
+ s->zfail_op != PIPE_STENCIL_OP_KEEP))
+ return TRUE;
+ return FALSE;
+}
+
+static boolean r300_can_hiz(struct r300_context *r300)
+{
+ struct r300_dsa_state *dsa_state = r300->dsa_state.state;
+ struct pipe_depth_stencil_alpha_state *dsa = &dsa_state->dsa;
+ struct r300_screen* r300screen = r300->screen;
+ struct r300_hyperz_state *z = r300->hyperz_state.state;
+
+ /* shader writes depth - no HiZ */
+ if (r300_fragment_shader_writes_depth(r300_fs(r300))) /* (5) */
+ return FALSE;
+
+ if (r300->query_current)
+ return FALSE;
+ /* if stencil fail/zfail op is not KEEP */
+ if (r300_dsa_stencil_op_not_keep(&dsa->stencil[0]) ||
+ r300_dsa_stencil_op_not_keep(&dsa->stencil[1]))
+ return FALSE;
+
+ if (dsa->depth.enabled) {
+ /* if depth func is EQUAL pre-r500 */
+ if (dsa->depth.func == PIPE_FUNC_EQUAL && !r300screen->caps.is_r500)
+ return FALSE;
+ /* if depth func is NOTEQUAL */
+ if (dsa->depth.func == PIPE_FUNC_NOTEQUAL)
+ return FALSE;
+ }
+ /* depth comparison function - if just cleared save and return okay */
+ if (z->current_func == -1) {
+ int func = dsa_state->z_stencil_control & 0x7;
+ if (func != 0 && func != 7)
+ z->current_func = dsa_state->z_stencil_control & 0x7;
+ } else {
+ /* simple don't change */
+ if (!r300_zfunc_same_direction(z->current_func, (dsa_state->z_stencil_control & 0x7))) {
+ DBG(r300, DBG_HYPERZ, "z func changed direction - disabling hyper-z %d -> %d\n", z->current_func, dsa_state->z_stencil_control);
+ return FALSE;
+ }
+ }
+ return TRUE;
+}
+
static void r300_update_hyperz(struct r300_context* r300)
{
struct r300_hyperz_state *z =
(struct r300_hyperz_state*)r300->hyperz_state.state;
+ z->gb_z_peq_config = 0;
z->zb_bw_cntl = 0;
z->sc_hyperz = R300_SC_HYPERZ_ADJ_2;
+ z->flush = 0;
- if (r300->cbzb_clear)
+ if (r300->cbzb_clear) {
z->zb_bw_cntl |= R300_ZB_CB_CLEAR_CACHE_LINE_WRITE_ONLY;
+ return;
+ }
+
+ /* Zbuffer compression. */
+ if (r300->z_compression) {
+ z->zb_bw_cntl |= R300_RD_COMP_ENABLE;
+ if (r300->z_decomp_rd == false)
+ z->zb_bw_cntl |= R300_WR_COMP_ENABLE;
+ /* RV350 and up optimizations. */
+ if (r300->z_compression == RV350_Z_COMPRESS_88)
+ z->gb_z_peq_config |= R300_GB_Z_PEQ_CONFIG_Z_PEQ_SIZE_8_8;
+ }
+
+ /* Z fastfill. */
+ if (r300->z_fastfill) {
+ z->zb_bw_cntl |= R300_FAST_FILL_ENABLE; /* | R300_FORCE_COMPRESSED_STENCIL_VALUE_ENABLE;*/
+ }
+
+ if (r300->hiz_enable) {
+ bool can_hiz = r300_can_hiz(r300);
+ if (can_hiz) {
+ z->zb_bw_cntl |= R300_HIZ_ENABLE;
+ z->sc_hyperz |= R300_SC_HYPERZ_ENABLE;
+ z->sc_hyperz |= r300_get_sc_hz_max(r300);
+ z->zb_bw_cntl |= r300_get_hiz_min(r300);
+ }
+ }
+
+ if (r300->screen->caps.is_r500) {
+ /* XXX Are these bits really available on RV350? */
+ z->zb_bw_cntl |= R500_HIZ_FP_EXP_BITS_3;
+ z->zb_bw_cntl |=
+ R500_HIZ_EQUAL_REJECT_ENABLE |
+ R500_PEQ_PACKING_ENABLE |
+ R500_COVERED_PTR_MASKING_ENABLE;
+ }
}
/*****************************************************************************/
@@ -126,15 +257,115 @@ static void r300_update_ztop(struct r300_context* r300)
} else {
ztop_state->z_buffer_top = R300_ZTOP_ENABLE;
}
-
if (ztop_state->z_buffer_top != old_ztop)
r300->ztop_state.dirty = TRUE;
}
+#define ALIGN_DIVUP(x, y) (((x) + (y) - 1) / (y))
+
+static void r300_update_hiz_clear(struct r300_context *r300)
+{
+ struct pipe_framebuffer_state *fb =
+ (struct pipe_framebuffer_state*)r300->fb_state.state;
+ uint32_t height;
+
+ height = ALIGN_DIVUP(fb->zsbuf->height, 4);
+ r300->hiz_clear.size = height * 4;
+}
+
+static void r300_update_zmask_clear(struct r300_context *r300)
+{
+ struct pipe_framebuffer_state *fb =
+ (struct pipe_framebuffer_state*)r300->fb_state.state;
+ uint32_t height;
+ int mult;
+
+ if (r300->z_compression == RV350_Z_COMPRESS_88)
+ mult = 8;
+ else
+ mult = 4;
+
+ height = ALIGN_DIVUP(fb->zsbuf->height, mult);
+
+ r300->zmask_clear.size = height * 4;
+}
+
void r300_update_hyperz_state(struct r300_context* r300)
{
r300_update_ztop(r300);
if (r300->hyperz_state.dirty) {
r300_update_hyperz(r300);
}
+
+ if (r300->hiz_clear.dirty) {
+ r300_update_hiz_clear(r300);
+ }
+ if (r300->zmask_clear.dirty) {
+ r300_update_zmask_clear(r300);
+ }
+}
+
+void r300_hiz_alloc_block(struct r300_context *r300, struct r300_surface *surf)
+{
+ struct r300_texture *tex;
+ uint32_t zsize, ndw;
+ int level = surf->base.level;
+
+ tex = r300_texture(surf->base.texture);
+
+ if (tex->hiz_mem[level])
+ return;
+
+ zsize = tex->desc.layer_size_in_bytes[level];
+ zsize /= util_format_get_blocksize(tex->desc.b.b.format);
+ ndw = ALIGN_DIVUP(zsize, 64);
+
+ tex->hiz_mem[level] = u_mmAllocMem(r300->hiz_mm, ndw, 0, 0);
+ return;
+}
+
+void r300_zmask_alloc_block(struct r300_context *r300, struct r300_surface *surf, int compress)
+{
+ int bsize = 256;
+ uint32_t zsize, ndw;
+ int level = surf->base.level;
+ struct r300_texture *tex;
+
+ tex = r300_texture(surf->base.texture);
+
+ if (tex->zmask_mem[level])
+ return;
+
+ zsize = tex->desc.layer_size_in_bytes[level];
+ zsize /= util_format_get_blocksize(tex->desc.b.b.format);
+
+ /* each zmask dword represents 16 4x4 blocks - which is 256 pixels
+ or 16 8x8 depending on the gb peq flag = 1024 pixels */
+ if (compress == RV350_Z_COMPRESS_88)
+ bsize = 1024;
+
+ ndw = ALIGN_DIVUP(zsize, bsize);
+ tex->zmask_mem[level] = u_mmAllocMem(r300->zmask_mm, ndw, 0, 0);
+ return;
+}
+
+void r300_hyperz_init_mm(struct r300_context *r300)
+{
+ struct r300_screen* r300screen = r300->screen;
+ int frag_pipes = r300screen->caps.num_frag_pipes;
+
+ if (r300screen->caps.hiz_ram)
+ r300->hiz_mm = u_mmInit(0, r300screen->caps.hiz_ram * frag_pipes);
+
+ r300->zmask_mm = u_mmInit(0, r300screen->caps.zmask_ram * frag_pipes);
+}
+
+void r300_hyperz_destroy_mm(struct r300_context *r300)
+{
+ struct r300_screen* r300screen = r300->screen;
+
+ if (r300screen->caps.hiz_ram)
+ u_mmDestroy(r300->hiz_mm);
+
+ u_mmDestroy(r300->zmask_mm);
}
diff --git a/src/gallium/drivers/r300/r300_hyperz.h b/src/gallium/drivers/r300/r300_hyperz.h
index 3df5053b896..09e1ff6625c 100644
--- a/src/gallium/drivers/r300/r300_hyperz.h
+++ b/src/gallium/drivers/r300/r300_hyperz.h
@@ -27,4 +27,9 @@ struct r300_context;
void r300_update_hyperz_state(struct r300_context* r300);
+void r300_hiz_alloc_block(struct r300_context *r300, struct r300_surface *surf);
+void r300_zmask_alloc_block(struct r300_context *r300, struct r300_surface *surf, int compress);
+
+void r300_hyperz_init_mm(struct r300_context *r300);
+void r300_hyperz_destroy_mm(struct r300_context *r300);
#endif
diff --git a/src/gallium/drivers/r300/r300_reg.h b/src/gallium/drivers/r300/r300_reg.h
index 2acc1a903e8..99a9d650551 100644
--- a/src/gallium/drivers/r300/r300_reg.h
+++ b/src/gallium/drivers/r300/r300_reg.h
@@ -3436,6 +3436,7 @@ enum {
# define R300_VBPNTR_SIZE1(x) (((x) >> 2) << 16)
# define R300_VBPNTR_STRIDE1(x) (((x) >> 2) << 24)
+#define R300_PACKET3_3D_CLEAR_ZMASK 0x00003200
#define R300_PACKET3_INDX_BUFFER 0x00003300
# define R300_INDX_BUFFER_DST_SHIFT 0
# define R300_INDX_BUFFER_SKIP_SHIFT 16
diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c
index bae02135da9..910f5f7113e 100644
--- a/src/gallium/drivers/r300/r300_render.c
+++ b/src/gallium/drivers/r300/r300_render.c
@@ -223,7 +223,8 @@ static void r300_prepare_for_rendering(struct r300_context *r300,
/* Emitted in flush. */
end_dwords += 26; /* emit_query_end */
- end_dwords += r300->hyperz_state.size; /* emit_hyperz_end */
+ if (r300->rws->get_value(r300->rws, R300_CAN_HYPERZ))
+ end_dwords += r300->hyperz_state.size + 2; /* emit_hyperz_end + zcache flush */
cs_dwords += end_dwords;
@@ -566,19 +567,6 @@ static void r300_draw_range_elements(struct pipe_context* pipe,
}
}
-/* Simple helpers for context setup. Should probably be moved to util. */
-static void r300_draw_elements(struct pipe_context* pipe,
- struct pipe_resource* indexBuffer,
- unsigned indexSize, int indexBias, unsigned mode,
- unsigned start, unsigned count)
-{
- struct r300_context *r300 = r300_context(pipe);
-
- pipe->draw_range_elements(pipe, indexBuffer, indexSize, indexBias,
- 0, r300->vertex_buffer_max_index,
- mode, start, count);
-}
-
static void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
unsigned start, unsigned count)
{
@@ -638,111 +626,104 @@ static void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
}
}
-/****************************************************************************
- * The rest of this file is for SW TCL rendering only. Please be polite and *
- * keep these functions separated so that they are easier to locate. ~C. *
- ***************************************************************************/
-
-/* SW TCL arrays, using Draw. */
-static void r300_swtcl_draw_arrays(struct pipe_context* pipe,
- unsigned mode,
- unsigned start,
- unsigned count)
+static void r300_draw_vbo(struct pipe_context* pipe,
+ const struct pipe_draw_info *info)
{
struct r300_context* r300 = r300_context(pipe);
- struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS];
- int i;
- if (r300->skip_rendering) {
- return;
+ if (info->indexed && r300->index_buffer.buffer) {
+ unsigned offset;
+
+ assert(r300->index_buffer.offset % r300->index_buffer.index_size == 0);
+ offset = r300->index_buffer.offset / r300->index_buffer.index_size;
+
+ r300_draw_range_elements(pipe,
+ r300->index_buffer.buffer,
+ r300->index_buffer.index_size,
+ info->index_bias,
+ info->min_index,
+ info->max_index,
+ info->mode,
+ info->start + offset,
+ info->count);
}
-
- if (!u_trim_pipe_prim(mode, &count)) {
- return;
- }
-
- r300_update_derived_state(r300);
-
- for (i = 0; i < r300->vertex_buffer_count; i++) {
- void* buf = pipe_buffer_map(pipe,
- r300->vertex_buffer[i].buffer,
- PIPE_TRANSFER_READ,
- &vb_transfer[i]);
- draw_set_mapped_vertex_buffer(r300->draw, i, buf);
- }
-
- draw_set_mapped_element_buffer(r300->draw, 0, 0, NULL);
-
- draw_arrays(r300->draw, mode, start, count);
-
- /* XXX Not sure whether this is the best fix.
- * It prevents CS from being rejected and weird assertion failures. */
- draw_flush(r300->draw);
-
- for (i = 0; i < r300->vertex_buffer_count; i++) {
- pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer,
- vb_transfer[i]);
- draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
+ else {
+ r300_draw_arrays(pipe,
+ info->mode,
+ info->start,
+ info->count);
}
}
+/****************************************************************************
+ * The rest of this file is for SW TCL rendering only. Please be polite and *
+ * keep these functions separated so that they are easier to locate. ~C. *
+ ***************************************************************************/
+
/* SW TCL elements, using Draw. */
-static void r300_swtcl_draw_range_elements(struct pipe_context* pipe,
- struct pipe_resource* indexBuffer,
- unsigned indexSize,
- int indexBias,
- unsigned minIndex,
- unsigned maxIndex,
- unsigned mode,
- unsigned start,
- unsigned count)
+static void r300_swtcl_draw_vbo(struct pipe_context* pipe,
+ const struct pipe_draw_info *info)
{
struct r300_context* r300 = r300_context(pipe);
struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS];
- struct pipe_transfer *ib_transfer;
+ struct pipe_transfer *ib_transfer = NULL;
+ unsigned count = info->count;
int i;
- void* indices;
+ void* indices = NULL;
if (r300->skip_rendering) {
return;
}
- if (!u_trim_pipe_prim(mode, &count)) {
+ if (!u_trim_pipe_prim(info->mode, &count)) {
return;
}
r300_update_derived_state(r300);
for (i = 0; i < r300->vertex_buffer_count; i++) {
- void* buf = pipe_buffer_map(pipe,
- r300->vertex_buffer[i].buffer,
- PIPE_TRANSFER_READ,
- &vb_transfer[i]);
- draw_set_mapped_vertex_buffer(r300->draw, i, buf);
+ if (r300->vertex_buffer[i].buffer) {
+ void *buf = pipe_buffer_map(pipe,
+ r300->vertex_buffer[i].buffer,
+ PIPE_TRANSFER_READ,
+ &vb_transfer[i]);
+ draw_set_mapped_vertex_buffer(r300->draw, i, buf);
+ }
+ }
+
+ if (info->indexed && r300->index_buffer.buffer) {
+ indices = pipe_buffer_map(pipe, r300->index_buffer.buffer,
+ PIPE_TRANSFER_READ, &ib_transfer);
+ if (indices)
+ indices = (void *) ((char *) indices + r300->index_buffer.offset);
}
- indices = pipe_buffer_map(pipe, indexBuffer,
- PIPE_TRANSFER_READ, &ib_transfer);
- draw_set_mapped_element_buffer_range(r300->draw, indexSize, indexBias,
- minIndex, maxIndex, indices);
+ draw_set_mapped_element_buffer_range(r300->draw, (indices) ?
+ r300->index_buffer.index_size : 0,
+ info->index_bias,
+ info->min_index,
+ info->max_index,
+ indices);
- draw_arrays(r300->draw, mode, start, count);
+ draw_arrays(r300->draw, info->mode, info->start, count);
/* XXX Not sure whether this is the best fix.
* It prevents CS from being rejected and weird assertion failures. */
draw_flush(r300->draw);
for (i = 0; i < r300->vertex_buffer_count; i++) {
- pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer,
- vb_transfer[i]);
- draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
+ if (r300->vertex_buffer[i].buffer) {
+ pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer,
+ vb_transfer[i]);
+ draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
+ }
}
- pipe_buffer_unmap(pipe, indexBuffer,
- ib_transfer);
- draw_set_mapped_element_buffer_range(r300->draw, 0, 0,
- start, start + count - 1,
- NULL);
+ if (ib_transfer) {
+ pipe_buffer_unmap(pipe, r300->index_buffer.buffer, ib_transfer);
+ draw_set_mapped_element_buffer_range(r300->draw, 0, 0, info->start,
+ info->start + count - 1, NULL);
+ }
}
/* Object for rendering using Draw. */
@@ -820,6 +801,8 @@ static void* r300_render_map_vertices(struct vbuf_render* render)
PIPE_TRANSFER_WRITE,
&r300render->vbo_transfer);
+ assert(r300render->vbo_ptr);
+
return ((uint8_t*)r300render->vbo_ptr + r300render->vbo_offset);
}
@@ -1141,16 +1124,11 @@ static void r300_resource_resolve(struct pipe_context* pipe,
void r300_init_render_functions(struct r300_context *r300)
{
- /* Set generic functions. */
- r300->context.draw_elements = r300_draw_elements;
-
/* Set draw functions based on presence of HW TCL. */
if (r300->screen->caps.has_tcl) {
- r300->context.draw_arrays = r300_draw_arrays;
- r300->context.draw_range_elements = r300_draw_range_elements;
+ r300->context.draw_vbo = r300_draw_vbo;
} else {
- r300->context.draw_arrays = r300_swtcl_draw_arrays;
- r300->context.draw_range_elements = r300_swtcl_draw_range_elements;
+ r300->context.draw_vbo = r300_swtcl_draw_vbo;
}
r300->context.resource_resolve = r300_resource_resolve;
diff --git a/src/gallium/drivers/r300/r300_render_stencilref.c b/src/gallium/drivers/r300/r300_render_stencilref.c
index 9a6b4e12ff1..1f035d64a28 100644
--- a/src/gallium/drivers/r300/r300_render_stencilref.c
+++ b/src/gallium/drivers/r300/r300_render_stencilref.c
@@ -34,13 +34,8 @@
#include "r300_reg.h"
struct r300_stencilref_context {
- void (*draw_arrays)(struct pipe_context *pipe,
- unsigned mode, unsigned start, unsigned count);
-
- void (*draw_range_elements)(
- struct pipe_context *pipe, struct pipe_resource *indexBuffer,
- unsigned indexSize, int indexBias, unsigned minIndex, unsigned maxIndex,
- unsigned mode, unsigned start, unsigned count);
+ void (*draw_vbo)(struct pipe_context *pipe,
+ const struct pipe_draw_info *info);
uint32_t rs_cull_mode;
uint32_t zb_stencilrefmask;
@@ -105,41 +100,19 @@ static void r300_stencilref_end(struct r300_context *r300)
r300->dsa_state.dirty = TRUE;
}
-static void r300_stencilref_draw_arrays(struct pipe_context *pipe, unsigned mode,
- unsigned start, unsigned count)
-{
- struct r300_context *r300 = r300_context(pipe);
- struct r300_stencilref_context *sr = r300->stencilref_fallback;
-
- if (!r300_stencilref_needed(r300)) {
- sr->draw_arrays(pipe, mode, start, count);
- } else {
- r300_stencilref_begin(r300);
- sr->draw_arrays(pipe, mode, start, count);
- r300_stencilref_switch_side(r300);
- sr->draw_arrays(pipe, mode, start, count);
- r300_stencilref_end(r300);
- }
-}
-
-static void r300_stencilref_draw_range_elements(
- struct pipe_context *pipe, struct pipe_resource *indexBuffer,
- unsigned indexSize, int indexBias, unsigned minIndex, unsigned maxIndex,
- unsigned mode, unsigned start, unsigned count)
+static void r300_stencilref_draw_vbo(struct pipe_context *pipe,
+ const struct pipe_draw_info *info)
{
struct r300_context *r300 = r300_context(pipe);
struct r300_stencilref_context *sr = r300->stencilref_fallback;
if (!r300_stencilref_needed(r300)) {
- sr->draw_range_elements(pipe, indexBuffer, indexSize, indexBias,
- minIndex, maxIndex, mode, start, count);
+ sr->draw_vbo(pipe, info);
} else {
r300_stencilref_begin(r300);
- sr->draw_range_elements(pipe, indexBuffer, indexSize, indexBias,
- minIndex, maxIndex, mode, start, count);
+ sr->draw_vbo(pipe, info);
r300_stencilref_switch_side(r300);
- sr->draw_range_elements(pipe, indexBuffer, indexSize, indexBias,
- minIndex, maxIndex, mode, start, count);
+ sr->draw_vbo(pipe, info);
r300_stencilref_end(r300);
}
}
@@ -148,11 +121,9 @@ void r300_plug_in_stencil_ref_fallback(struct r300_context *r300)
{
r300->stencilref_fallback = CALLOC_STRUCT(r300_stencilref_context);
- /* Save original draw functions. */
- r300->stencilref_fallback->draw_arrays = r300->context.draw_arrays;
- r300->stencilref_fallback->draw_range_elements = r300->context.draw_range_elements;
+ /* Save original draw function. */
+ r300->stencilref_fallback->draw_vbo = r300->context.draw_vbo;
- /* Override the draw functions. */
- r300->context.draw_arrays = r300_stencilref_draw_arrays;
- r300->context.draw_range_elements = r300_stencilref_draw_range_elements;
+ /* Override the draw function. */
+ r300->context.draw_vbo = r300_stencilref_draw_vbo;
}
diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c
index 676430f5fee..6268001054b 100644
--- a/src/gallium/drivers/r300/r300_screen.c
+++ b/src/gallium/drivers/r300/r300_screen.c
@@ -279,11 +279,14 @@ static boolean r300_is_format_supported(struct pipe_screen* screen,
case 3:
case 4:
case 6:
+ return FALSE;
+#if 0
if (usage != PIPE_BIND_RENDER_TARGET ||
!util_format_is_rgba8_variant(
util_format_description(format))) {
return FALSE;
}
+#endif
break;
default:
return FALSE;
diff --git a/src/gallium/drivers/r300/r300_screen.h b/src/gallium/drivers/r300/r300_screen.h
index 18745b83a09..13a3320b992 100644
--- a/src/gallium/drivers/r300/r300_screen.h
+++ b/src/gallium/drivers/r300/r300_screen.h
@@ -91,6 +91,7 @@ r300_winsys_screen(struct pipe_screen *screen) {
#define DBG_FB (1 << 9)
#define DBG_RS_BLOCK (1 << 10)
#define DBG_CBZB (1 << 11)
+#define DBG_HYPERZ (1 << 12)
/* Features. */
#define DBG_ANISOHQ (1 << 16)
#define DBG_NO_TILING (1 << 17)
diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index 3e221f2e02d..1e6b81d7989 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. */
@@ -446,7 +448,6 @@ static void r300_set_clip_state(struct pipe_context* pipe,
r300->clip_state.dirty = TRUE;
} else {
- draw_flush(r300->draw);
draw_set_clip_state(r300->draw, state);
}
}
@@ -473,14 +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);
@@ -593,6 +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);
}
@@ -682,11 +684,13 @@ void r300_mark_fb_state_dirty(struct r300_context *r300,
enum r300_fb_state_change change)
{
struct pipe_framebuffer_state *state = r300->fb_state.state;
+ boolean has_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
/* 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;
@@ -698,8 +702,11 @@ 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;
+ else if (state->zsbuf) {
+ r300->fb_state.size += 10;
+ if (has_hyperz)
+ r300->fb_state.size += r300->screen->caps.hiz_ram ? 8 : 4;
+ }
/* The size of the rest of atoms stays the same. */
}
@@ -711,8 +718,10 @@ 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;
@@ -728,10 +737,6 @@ static void
return;
}
- if (r300->draw) {
- draw_flush(r300->draw);
- }
-
/* If nr_cbufs is changed from zero to non-zero or vice versa... */
if (!!old_state->nr_cbufs != !!state->nr_cbufs) {
r300->blend_state.dirty = TRUE;
@@ -748,20 +753,57 @@ static void
r300_mark_fb_state_dirty(r300, R300_CHANGED_FB_STATE);
- /* Polygon offset depends on the zbuffer bit depth. */
- if (state->zsbuf && r300->polygon_offset_enabled) {
- switch (util_format_get_blocksize(state->zsbuf->texture->format)) {
- case 2:
- zbuffer_bpp = 16;
- break;
- case 4:
- zbuffer_bpp = 24;
- break;
+ r300->hiz_enable = false;
+ r300->z_fastfill = false;
+ r300->z_compression = false;
+
+ if (state->zsbuf) {
+ 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;
- r300->rs_state.dirty = TRUE;
+
+ if (r300->polygon_offset_enabled)
+ r300->rs_state.dirty = TRUE;
}
}
@@ -1019,7 +1061,7 @@ static void* r300_create_rs_state(struct pipe_context* pipe,
for (i = 0; i < 8; i++) {
if (state->sprite_coord_enable & (1 << i))
stuffing_enable |=
- R300_GB_TEX_STR << (R300_GB_TEX0_SOURCE_SHIFT + (i*2));
+ R300_GB_TEX_ST << (R300_GB_TEX0_SOURCE_SHIFT + (i*2));
}
point_texcoord_left = 0.0f;
@@ -1096,14 +1138,11 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
boolean last_two_sided_color = r300->two_sided_color;
if (r300->draw && rs) {
- draw_flush(r300->draw);
draw_set_rasterizer_state(r300->draw, &rs->rs_draw, state);
}
if (rs) {
- r300->polygon_offset_enabled = (rs->rs.offset_point ||
- rs->rs.offset_line ||
- rs->rs.offset_tri);
+ r300->polygon_offset_enabled = rs->polygon_offset_enable;
r300->sprite_coord_enable = rs->rs.sprite_coord_enable;
r300->two_sided_color = rs->rs.light_twoside;
} else {
@@ -1385,7 +1424,6 @@ static void r300_set_viewport_state(struct pipe_context* pipe,
r300->viewport = *state;
if (r300->draw) {
- draw_flush(r300->draw);
draw_set_viewport_state(r300->draw, state);
viewport->vte_control = R300_VTX_XY_FMT | R300_VTX_Z_FMT;
return;
@@ -1486,7 +1524,6 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe,
} else {
/* SW TCL. */
- draw_flush(r300->draw);
draw_set_vertex_buffers(r300->draw, count, buffers);
}
@@ -1505,6 +1542,23 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe,
r300->vertex_buffer_count = count;
}
+static void r300_set_index_buffer(struct pipe_context* pipe,
+ const struct pipe_index_buffer *ib)
+{
+ struct r300_context* r300 = r300_context(pipe);
+
+ if (ib) {
+ pipe_resource_reference(&r300->index_buffer.buffer, ib->buffer);
+ memcpy(&r300->index_buffer, ib, sizeof(r300->index_buffer));
+ }
+ else {
+ pipe_resource_reference(&r300->index_buffer.buffer, NULL);
+ memset(&r300->index_buffer, 0, sizeof(r300->index_buffer));
+ }
+
+ /* TODO make this more like a state */
+}
+
/* Initialize the PSC tables. */
static void r300_vertex_psc(struct r300_vertex_element_state *velems)
{
@@ -1654,7 +1708,6 @@ static void r300_bind_vertex_elements_state(struct pipe_context *pipe,
r300->velems = velems;
if (r300->draw) {
- draw_flush(r300->draw);
draw_set_vertex_elements(r300->draw, velems->count, velems->velem);
return;
}
@@ -1720,7 +1773,6 @@ static void r300_bind_vs_state(struct pipe_context* pipe, void* shader)
r300->pvs_flush.dirty = TRUE;
} else {
- draw_flush(r300->draw);
draw_bind_vertex_shader(r300->draw,
(struct draw_vertex_shader*)vs->draw_vs);
}
@@ -1852,6 +1904,7 @@ void r300_init_state_functions(struct r300_context* r300)
r300->context.set_viewport_state = r300_set_viewport_state;
r300->context.set_vertex_buffers = r300_set_vertex_buffers;
+ r300->context.set_index_buffer = r300_set_index_buffer;
r300->context.create_vertex_elements_state = r300_create_vertex_elements_state;
r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state;
diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c
index a85db27064c..f3dad4c2923 100644
--- a/src/gallium/drivers/r300/r300_state_derived.c
+++ b/src/gallium/drivers/r300/r300_state_derived.c
@@ -35,6 +35,7 @@
#include "r300_state_inlines.h"
#include "r300_texture.h"
#include "r300_vs.h"
+#include "r300_winsys.h"
/* r300_state_derived: Various bits of state which are dependent upon
* currently bound CSO data. */
@@ -213,19 +214,19 @@ static void r300_rs_tex(struct r300_rs_block* rs, int id, int ptr,
enum r300_rs_swizzle swiz)
{
if (swiz == SWIZ_X001) {
- rs->ip[id] |= R300_RS_TEX_PTR(ptr*4) |
+ rs->ip[id] |= R300_RS_TEX_PTR(ptr) |
R300_RS_SEL_S(R300_RS_SEL_C0) |
R300_RS_SEL_T(R300_RS_SEL_K0) |
R300_RS_SEL_R(R300_RS_SEL_K0) |
R300_RS_SEL_Q(R300_RS_SEL_K1);
} else if (swiz == SWIZ_XY01) {
- rs->ip[id] |= R300_RS_TEX_PTR(ptr*4) |
+ rs->ip[id] |= R300_RS_TEX_PTR(ptr) |
R300_RS_SEL_S(R300_RS_SEL_C0) |
R300_RS_SEL_T(R300_RS_SEL_C1) |
R300_RS_SEL_R(R300_RS_SEL_K0) |
R300_RS_SEL_Q(R300_RS_SEL_K1);
} else {
- rs->ip[id] |= R300_RS_TEX_PTR(ptr*4) |
+ rs->ip[id] |= R300_RS_TEX_PTR(ptr) |
R300_RS_SEL_S(R300_RS_SEL_C0) |
R300_RS_SEL_T(R300_RS_SEL_C1) |
R300_RS_SEL_R(R300_RS_SEL_C2) |
@@ -261,23 +262,21 @@ static void r500_rs_col_write(struct r300_rs_block* rs, int id, int fp_offset)
static void r500_rs_tex(struct r300_rs_block* rs, int id, int ptr,
enum r300_rs_swizzle swiz)
{
- int rs_tex_comp = ptr*4;
-
if (swiz == SWIZ_X001) {
- rs->ip[id] |= R500_RS_SEL_S(rs_tex_comp) |
+ rs->ip[id] |= R500_RS_SEL_S(ptr) |
R500_RS_SEL_T(R500_RS_IP_PTR_K0) |
R500_RS_SEL_R(R500_RS_IP_PTR_K0) |
R500_RS_SEL_Q(R500_RS_IP_PTR_K1);
} else if (swiz == SWIZ_XY01) {
- rs->ip[id] |= R500_RS_SEL_S(rs_tex_comp) |
- R500_RS_SEL_T(rs_tex_comp + 1) |
+ rs->ip[id] |= R500_RS_SEL_S(ptr) |
+ R500_RS_SEL_T(ptr + 1) |
R500_RS_SEL_R(R500_RS_IP_PTR_K0) |
R500_RS_SEL_Q(R500_RS_IP_PTR_K1);
} else {
- rs->ip[id] |= R500_RS_SEL_S(rs_tex_comp) |
- R500_RS_SEL_T(rs_tex_comp + 1) |
- R500_RS_SEL_R(rs_tex_comp + 2) |
- R500_RS_SEL_Q(rs_tex_comp + 3);
+ rs->ip[id] |= R500_RS_SEL_S(ptr) |
+ R500_RS_SEL_T(ptr + 1) |
+ R500_RS_SEL_R(ptr + 2) |
+ R500_RS_SEL_Q(ptr + 3);
}
rs->inst[id] |= R500_RS_INST_TEX_ID(id);
}
@@ -305,7 +304,7 @@ static void r300_update_rs_block(struct r300_context *r300)
struct r300_shader_semantics *vs_outputs = &vs->outputs;
struct r300_shader_semantics *fs_inputs = &r300_fs(r300)->shader->inputs;
struct r300_rs_block rs = {0};
- int i, col_count = 0, tex_count = 0, fp_offset = 0, count, loc = 0;
+ int i, col_count = 0, tex_count = 0, fp_offset = 0, count, loc = 0, tex_ptr = 0;
void (*rX00_rs_col)(struct r300_rs_block*, int, int, enum r300_rs_swizzle);
void (*rX00_rs_col_write)(struct r300_rs_block*, int, int);
void (*rX00_rs_tex)(struct r300_rs_block*, int, int, enum r300_rs_swizzle);
@@ -393,8 +392,9 @@ static void r300_update_rs_block(struct r300_context *r300)
stream_loc_notcl[loc++] = 6 + tex_count;
/* Rasterize it. */
- rX00_rs_tex(&rs, tex_count, tex_count, SWIZ_XYZW);
+ rX00_rs_tex(&rs, tex_count, tex_ptr, SWIZ_XYZW);
tex_count++;
+ tex_ptr += 4;
}
}
}
@@ -412,7 +412,7 @@ static void r300_update_rs_block(struct r300_context *r300)
}
/* Rasterize it. */
- rX00_rs_tex(&rs, tex_count, tex_count,
+ rX00_rs_tex(&rs, tex_count, tex_ptr,
sprite_coord ? SWIZ_XY01 : SWIZ_XYZW);
/* Write it to the FS input register if it's needed by the FS. */
@@ -429,6 +429,7 @@ static void r300_update_rs_block(struct r300_context *r300)
i, sprite_coord ? " (sprite coord)" : "");
}
tex_count++;
+ tex_ptr += sprite_coord ? 2 : 4;
} else {
/* Skip the FS input register, leave it uninitialized. */
/* If we try to set it to (0,0,0,1), it will lock up. */
@@ -449,7 +450,7 @@ static void r300_update_rs_block(struct r300_context *r300)
stream_loc_notcl[loc++] = 6 + tex_count;
/* Rasterize it. */
- rX00_rs_tex(&rs, tex_count, tex_count, SWIZ_X001);
+ rX00_rs_tex(&rs, tex_count, tex_ptr, SWIZ_X001);
/* Write it to the FS input register if it's needed by the FS. */
if (fs_inputs->fog != ATTR_UNUSED) {
@@ -461,6 +462,7 @@ static void r300_update_rs_block(struct r300_context *r300)
DBG(r300, DBG_RS, "r300: Rasterized fog unused.\n");
}
tex_count++;
+ tex_ptr += 4;
} else {
/* Skip the FS input register, leave it uninitialized. */
/* If we try to set it to (0,0,0,1), it will lock up. */
@@ -480,7 +482,7 @@ static void r300_update_rs_block(struct r300_context *r300)
stream_loc_notcl[loc++] = 6 + tex_count;
/* Rasterize it. */
- rX00_rs_tex(&rs, tex_count, tex_count, SWIZ_XYZW);
+ rX00_rs_tex(&rs, tex_count, tex_ptr, SWIZ_XYZW);
/* Write it to the FS input register. */
rX00_rs_tex_write(&rs, tex_count, fp_offset);
@@ -489,6 +491,7 @@ static void r300_update_rs_block(struct r300_context *r300)
fp_offset++;
tex_count++;
+ tex_ptr += 4;
}
/* Invalidate the rest of the no-TCL (GA) stream locations. */
@@ -507,7 +510,7 @@ static void r300_update_rs_block(struct r300_context *r300)
DBG(r300, DBG_RS, "r300: --- Rasterizer status ---: colors: %i, "
"generics: %i.\n", col_count, tex_count);
- rs.count = (tex_count*4) | (col_count << R300_IC_COUNT_SHIFT) |
+ rs.count = MIN2(tex_ptr, 32) | (col_count << R300_IC_COUNT_SHIFT) |
R300_HIRES_EN;
count = MAX3(col_count, tex_count, 1);
@@ -691,5 +694,6 @@ void r300_update_derived_state(struct r300_context* r300)
}
}
- r300_update_hyperz_state(r300);
+ if (r300->rws->get_value(r300->rws, R300_CAN_HYPERZ))
+ r300_update_hyperz_state(r300);
}
diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c
index fcdca5605e9..da8eadd3b53 100644
--- a/src/gallium/drivers/r300/r300_texture.c
+++ b/src/gallium/drivers/r300/r300_texture.c
@@ -35,6 +35,7 @@
#include "util/u_format_s3tc.h"
#include "util/u_math.h"
#include "util/u_memory.h"
+#include "util/u_mm.h"
#include "pipe/p_screen.h"
@@ -645,8 +646,16 @@ static void r300_texture_destroy(struct pipe_screen *screen,
{
struct r300_texture* tex = (struct r300_texture*)texture;
struct r300_winsys_screen *rws = (struct r300_winsys_screen *)texture->screen->winsys;
+ int i;
rws->buffer_reference(rws, &tex->buffer, NULL);
+ for (i = 0; i < R300_MAX_TEXTURE_LEVELS; i++) {
+ if (tex->hiz_mem[i])
+ u_mmFreeMem(tex->hiz_mem[i]);
+ if (tex->zmask_mem[i])
+ u_mmFreeMem(tex->zmask_mem[i]);
+ }
+
FREE(tex);
}
diff --git a/src/gallium/drivers/r300/r300_texture_desc.c b/src/gallium/drivers/r300/r300_texture_desc.c
index 343089bf2c5..5d690e8c332 100644
--- a/src/gallium/drivers/r300/r300_texture_desc.c
+++ b/src/gallium/drivers/r300/r300_texture_desc.c
@@ -136,13 +136,27 @@ static unsigned r300_texture_get_stride(struct r300_screen *screen,
stride = util_format_get_stride(desc->b.b.format, width);
- /* Some IGPs need a minimum stride of 64 bytes, hmm...
- * This doesn't seem to apply to tiled textures, according to r300c. */
- if (!desc->microtile && !desc->macrotile[level] &&
+ /* Some IGPs need a minimum stride of 64 bytes, hmm... */
+ if (!desc->macrotile[level] &&
(screen->caps.family == CHIP_FAMILY_RS600 ||
screen->caps.family == CHIP_FAMILY_RS690 ||
screen->caps.family == CHIP_FAMILY_RS740)) {
- return stride < 64 ? 64 : stride;
+ unsigned min_stride;
+
+ if (desc->microtile) {
+ unsigned tile_height =
+ r300_get_pixel_alignment(desc->b.b.format,
+ desc->b.b.nr_samples,
+ desc->microtile,
+ desc->macrotile[level],
+ DIM_HEIGHT);
+
+ min_stride = 64 / tile_height;
+ } else {
+ min_stride = 64;
+ }
+
+ return stride < min_stride ? min_stride : stride;
}
/* The alignment to 32 bytes is sort of implied by the layout... */
diff --git a/src/gallium/drivers/r300/r300_winsys.h b/src/gallium/drivers/r300/r300_winsys.h
index ff11546a647..187780750fa 100644
--- a/src/gallium/drivers/r300/r300_winsys.h
+++ b/src/gallium/drivers/r300/r300_winsys.h
@@ -49,6 +49,8 @@ enum r300_value_id {
R300_VID_Z_PIPES,
R300_VID_SQUARE_TILING_SUPPORT,
R300_VID_DRM_2_3_0,
+ R300_VID_DRM_2_6_0,
+ R300_CAN_HYPERZ,
};
enum r300_reference_domain { /* bitfield */