r300g: fixup arb occulsion query support.

1: add rv530 support
   - num z pipes cap
   - add proper start/finish query options for rv530

2: convert to use linked list properly.

3: add flushing required check.

4: initial Z top disabling support.

TODO:
make it actually work on my rv530.

11 files changed, 134 insertions, 43 deletions

diff --git a/src/gallium/drivers/r300/r300_chipset.h b/src/gallium/drivers/r300/r300_chipset.h
index 322d4a57e4..f015a4243d 100644
--- a/src/gallium/drivers/r300/r300_chipset.h
+++ b/src/gallium/drivers/r300/r300_chipset.h
@@ -36,6 +36,8 @@ struct r300_capabilities {
     int num_vert_fpus;
     /* The number of fragment pipes */
     int num_frag_pipes;
+    /* The number of z pipes */
+    int num_z_pipes;
     /* Whether or not TCL is physically present */
     boolean has_tcl;
     /* Whether or not this is an RV515 or newer; R500s have many differences
diff --git a/src/gallium/drivers/r300/r300_context.c b/src/gallium/drivers/r300/r300_context.c
index e6bc80e48f..7a9c098e30 100644
--- a/src/gallium/drivers/r300/r300_context.c
+++ b/src/gallium/drivers/r300/r300_context.c
@@ -99,11 +99,9 @@ static void r300_destroy_context(struct pipe_context* context) {
     context->screen->buffer_destroy(r300->oqbo);
 
     /* If there are any queries pending or not destroyed, remove them now. */
-    if (r300->query_list) {
-        foreach_s(query, temp, r300->query_list) {
-            remove_from_list(query);
-            FREE(query);
-        }
+    foreach_s(query, temp, &r300->query_list) {
+        remove_from_list(query);
+        FREE(query);
     }
 
     FREE(r300->blend_color_state);
@@ -201,6 +199,6 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
     r300->winsys->set_flush_cb(r300->winsys, r300_flush_cb, r300);
     r300->dirty_state = R300_NEW_KITCHEN_SINK;
     r300->dirty_hw++;
-
+    make_empty_list(&r300->query_list);
     return &r300->context;
 }
diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h
index 086633f732..9b0094b63c 100644
--- a/src/gallium/drivers/r300/r300_context.h
+++ b/src/gallium/drivers/r300/r300_context.h
@@ -172,6 +172,8 @@ struct r300_query {
     unsigned int count;
     /* The offset of this query into the query buffer, in bytes. */
     unsigned offset;
+    /* if we've flushed the query */
+    boolean flushed;
     /* Linked list members. */
     struct r300_query* prev;
     struct r300_query* next;
@@ -237,7 +239,7 @@ struct r300_context {
     /* Occlusion query buffer. */
     struct pipe_buffer* oqbo;
     /* Query list. */
-    struct r300_query* query_list;
+    struct r300_query query_list;
 
     /* Various CSO state objects. */
     /* Blend state. */
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index 64748ad8f8..3d28249c16 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -323,28 +323,30 @@ void r300_emit_fb_state(struct r300_context* r300,
 void r300_emit_query_begin(struct r300_context* r300,
                            struct r300_query* query)
 {
+    struct r300_capabilities* caps = r300_screen(r300->context.screen)->caps;
     CS_LOCALS(r300);
 
     /* XXX This will almost certainly not return good results
      * for overlapping queries. */
-    BEGIN_CS(2);
+    BEGIN_CS(4);
+    if (caps->family == CHIP_FAMILY_RV530) {
+	OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL);
+    } else {
+	OUT_CS_REG(R300_SU_REG_DEST, R300_RASTER_PIPE_SELECT_ALL);
+    }
     OUT_CS_REG(R300_ZB_ZPASS_DATA, 0);
     END_CS;
 }
 
-void r300_emit_query_end(struct r300_context* r300,
-                         struct r300_query* query)
+
+static void r300_emit_query_finish(struct r300_context *r300,
+				   struct r300_query *query)
 {
     struct r300_capabilities* caps = r300_screen(r300->context.screen)->caps;
     CS_LOCALS(r300);
 
-    if (!r300->winsys->add_buffer(r300->winsys, r300->oqbo,
-                0, RADEON_GEM_DOMAIN_GTT)) {
-        debug_printf("r300: There wasn't room for the OQ buffer!?"
-                " Oh noes!\n");
-    }
-
     assert(caps->num_frag_pipes);
+
     BEGIN_CS(6 * caps->num_frag_pipes + 2);
     /* I'm not so sure I like this switch, but it's hard to be elegant
      * when there's so many special cases...
@@ -394,6 +396,55 @@ void r300_emit_query_end(struct r300_context* r300,
 
 }
 
+static void rv530_emit_query_single(struct r300_context *r300,
+				    struct r300_query *query)
+{
+    CS_LOCALS(r300);
+
+    BEGIN_CS(8);
+    OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_0);
+    OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1);
+    OUT_CS_RELOC(r300->oqbo, query->offset, 0, RADEON_GEM_DOMAIN_GTT, 0);
+    OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL);
+    END_CS;
+}
+
+static void rv530_emit_query_double(struct r300_context *r300,
+				    struct r300_query *query)
+{
+    CS_LOCALS(r300);
+
+    BEGIN_CS(14);
+    OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_0);
+    OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1);
+    OUT_CS_RELOC(r300->oqbo, query->offset, 0, RADEON_GEM_DOMAIN_GTT, 0);
+    OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_1);
+    OUT_CS_REG_SEQ(R300_ZB_ZPASS_ADDR, 1);
+    OUT_CS_RELOC(r300->oqbo, query->offset + sizeof(uint32_t), 0, RADEON_GEM_DOMAIN_GTT, 0);
+    OUT_CS_REG(RV530_FG_ZBREG_DEST, RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL);
+    END_CS;
+}
+
+void r300_emit_query_end(struct r300_context* r300,
+                         struct r300_query* query)
+{
+    struct r300_capabilities *caps = r300_screen(r300->context.screen)->caps;
+
+    if (!r300->winsys->add_buffer(r300->winsys, r300->oqbo,
+                0, RADEON_GEM_DOMAIN_GTT)) {
+        debug_printf("r300: There wasn't room for the OQ buffer!?"
+                " Oh noes!\n");
+    }
+
+    if (caps->family == CHIP_FAMILY_RV530) {
+	if (caps->num_z_pipes == 2)
+	    rv530_emit_query_double(r300, query);
+	else
+	    rv530_emit_query_single(r300, query);
+    } else 
+        r300_emit_query_finish(r300, query);
+}
+
 void r300_emit_rs_state(struct r300_context* r300, struct r300_rs_state* rs)
 {
     CS_LOCALS(r300);
diff --git a/src/gallium/drivers/r300/r300_flush.c b/src/gallium/drivers/r300/r300_flush.c
index 0dff1c6f4f..a8ab0d7212 100644
--- a/src/gallium/drivers/r300/r300_flush.c
+++ b/src/gallium/drivers/r300/r300_flush.c
@@ -26,9 +26,10 @@ static void r300_flush(struct pipe_context* pipe,
                        unsigned flags,
                        struct pipe_fence_handle** fence)
 {
-    struct r300_context* r300 = r300_context(pipe);
-    CS_LOCALS(r300);
+    struct r300_context *r300 = r300_context(pipe);
+    struct r300_query *query;
 
+    CS_LOCALS(r300);
     /* 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. */
     if (!r300->draw->flushing) {
@@ -41,8 +42,13 @@ static void r300_flush(struct pipe_context* pipe,
         r300->dirty_state = R300_NEW_KITCHEN_SINK;
         r300->dirty_hw = 0;
     }
+    /* reset flushed query */
+    foreach(query, &r300->query_list) {
+        query->flushed = TRUE;
+    }
 }
 
+
 void r300_init_flush_functions(struct r300_context* r300)
 {
     r300->context.flush = r300_flush;
diff --git a/src/gallium/drivers/r300/r300_query.c b/src/gallium/drivers/r300/r300_query.c
index 2880d34877..b01313648b 100644
--- a/src/gallium/drivers/r300/r300_query.c
+++ b/src/gallium/drivers/r300/r300_query.c
@@ -24,13 +24,13 @@
 
 #include "r300_emit.h"
 
-static struct pipe_query* r300_create_query(struct pipe_context* pipe,
+static struct pipe_query *r300_create_query(struct pipe_context *pipe,
                                             unsigned query_type)
 {
-    struct r300_context* r300 = r300_context(pipe);
-    struct r300_screen* r300screen = r300_screen(r300->context.screen);
-    unsigned query_size = r300screen->caps->num_frag_pipes * 4;
-    struct r300_query* q, * qptr;
+    struct r300_context *r300 = r300_context(pipe);
+    struct r300_screen *r300screen = r300_screen(r300->context.screen);
+    unsigned query_size;
+    struct r300_query *q, *qptr;
 
     q = CALLOC_STRUCT(r300_query);
 
@@ -39,13 +39,16 @@ static struct pipe_query* r300_create_query(struct pipe_context* pipe,
 
     q->active = FALSE;
 
-    if (!r300->query_list) {
-        r300->query_list = q;
-    } else if (!is_empty_list(r300->query_list)) {
-        qptr = last_elem(r300->query_list);
+    if (r300screen->caps->family == CHIP_FAMILY_RV530)
+	query_size = r300screen->caps->num_z_pipes * sizeof(uint32_t);
+    else
+	query_size = r300screen->caps->num_frag_pipes * sizeof(uint32_t);
+
+    if (!is_empty_list(&r300->query_list)) {
+        qptr = last_elem(&r300->query_list);
         q->offset = qptr->offset + query_size;
-        insert_at_tail(r300->query_list, q);
     }
+    insert_at_tail(&r300->query_list, q);
 
     /* XXX */
     if (q->offset >= 4096) {
@@ -74,9 +77,10 @@ static void r300_begin_query(struct pipe_context* pipe,
     map = pipe->screen->buffer_map(pipe->screen, r300->oqbo,
             PIPE_BUFFER_USAGE_CPU_WRITE);
     map += q->offset / 4;
-    *map = ~0;
+    *map = ~0U;
     pipe->screen->buffer_unmap(pipe->screen, r300->oqbo);
 
+    q->flushed = FALSE;
     r300_emit_dirty_state(r300);
     r300_emit_query_begin(r300, q);
 }
@@ -98,28 +102,30 @@ static boolean r300_get_query_result(struct pipe_context* pipe,
 {
     struct r300_context* r300 = r300_context(pipe);
     struct r300_screen* r300screen = r300_screen(r300->context.screen);
-    struct r300_query* q = (struct r300_query*)query;
+    struct r300_query *q = (struct r300_query*)query;
     unsigned flags = PIPE_BUFFER_USAGE_CPU_READ;
     uint32_t* map;
-    uint32_t temp;
+    uint32_t temp = 0;
     unsigned i;
 
-    if (wait) {
+    if (q->flushed == FALSE)
         pipe->flush(pipe, 0, NULL);
-    } else {
+    if (!wait) {
         flags |= PIPE_BUFFER_USAGE_DONTBLOCK;
     }
 
     map = pipe->screen->buffer_map(pipe->screen, r300->oqbo, flags);
+    if (!map)
+        return FALSE;
     map += q->offset / 4;
     for (i = 0; i < r300screen->caps->num_frag_pipes; i++) {
-        if (*map == ~0) {
+        if (*map == ~0U) {
             /* Looks like our results aren't ready yet. */
             if (wait) {
                 debug_printf("r300: Despite waiting, OQ results haven't"
                         " come in yet.\n");
             }
-            temp = ~0;
+            temp = ~0U;
             break;
         }
         temp += *map;
@@ -127,7 +133,7 @@ static boolean r300_get_query_result(struct pipe_context* pipe,
     }
     pipe->screen->buffer_unmap(pipe->screen, r300->oqbo);
 
-    if (temp == ~0) {
+    if (temp == ~0U) {
         /* Our results haven't been written yet... */
         return FALSE;
     }
diff --git a/src/gallium/drivers/r300/r300_reg.h b/src/gallium/drivers/r300/r300_reg.h
index 3abff5db62..ae94bb9b9f 100644
--- a/src/gallium/drivers/r300/r300_reg.h
+++ b/src/gallium/drivers/r300/r300_reg.h
@@ -1172,6 +1172,13 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 /* SU Depth Offset value */
 #define R300_SU_DEPTH_OFFSET                0x42c4
 
+#define R300_SU_REG_DEST		    0x42c8
+#	define R300_RASTER_PIPE_SELECT_0	(1 << 0)
+#	define R300_RASTER_PIPE_SELECT_1	(1 << 1)
+#	define R300_RASTER_PIPE_SELECT_2	(1 << 2)
+#	define R300_RASTER_PIPE_SELECT_3	(1 << 3)
+#	define R300_RASTER_PIPE_SELECT_ALL	0xf
+
 
 /* BEGIN: Rasterization / Interpolators - many guesses */
 
@@ -2095,6 +2102,10 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #define R500_FG_ALPHA_VALUE                0x4be0
 #	define R500_FG_ALPHA_VALUE_MASK 0x0000ffff
 
+#define RV530_FG_ZBREG_DEST                 0x4be8
+#	define RV530_FG_ZBREG_DEST_PIPE_SELECT_0             (1 << 0)
+#	define RV530_FG_ZBREG_DEST_PIPE_SELECT_1             (1 << 1)
+#	define RV530_FG_ZBREG_DEST_PIPE_SELECT_ALL           (3 << 0)
 /* gap */
 
 /* Fragment program parameters in 7.16 floating point */
@@ -3313,10 +3324,6 @@ enum {
 
 #define R200_3D_DRAW_IMMD_2      0xC0003500
 
-/* XXX Oh look, stuff not brought over from docs yet */
-
-#define R300_SU_REG_DEST                    0x42C8
-
 #endif /* _R300_REG_H */
 
 /* *INDENT-ON* */
diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c
index 7d154576e0..5381651c77 100644
--- a/src/gallium/drivers/r300/r300_screen.c
+++ b/src/gallium/drivers/r300/r300_screen.c
@@ -395,6 +395,7 @@ struct pipe_screen* r300_create_screen(struct r300_winsys* r300_winsys)
 
     caps->pci_id = r300_winsys->pci_id;
     caps->num_frag_pipes = r300_winsys->gb_pipes;
+    caps->num_z_pipes = r300_winsys->z_pipes;
 
     r300_parse_chipset(caps);
 
diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index 3cef285dee..d8533ac168 100644
--- a/src/gallium/drivers/r300/r300_state.c
+++ b/src/gallium/drivers/r300/r300_state.c
@@ -190,6 +190,7 @@ static void*
         r300_create_dsa_state(struct pipe_context* pipe,
                               const struct pipe_depth_stencil_alpha_state* state)
 {
+    struct r300_context* r300 = r300_context(pipe);
     struct r300_dsa_state* dsa = CALLOC_STRUCT(r300_dsa_state);
 
     /* Depth test setup. */
@@ -247,11 +248,15 @@ static void*
             R300_FG_ALPHA_FUNC_ENABLE;
         dsa->alpha_reference = CLAMP(state->alpha.ref_value * 1023.0f,
                                      0, 1023);
-    } else {
-        /* XXX need to fix this to be dynamically set
-        dsa->z_buffer_top = R300_ZTOP_ENABLE; */
     }
 
+    dsa->z_buffer_top = R300_ZTOP_ENABLE;
+    /* XXX TODO: add frag prog rules for ztop disable */
+    if (state->alpha.enabled && state->alpha.func != PIPE_FUNC_ALWAYS)
+	dsa->z_buffer_top = R300_ZTOP_DISABLE;
+    if (!is_empty_list(&r300->query_list))
+	dsa->z_buffer_top = R300_ZTOP_DISABLE;
+
     return (void*)dsa;
 }
 
diff --git a/src/gallium/drivers/r300/r300_winsys.h b/src/gallium/drivers/r300/r300_winsys.h
index 540f8eca92..864a6146b2 100644
--- a/src/gallium/drivers/r300/r300_winsys.h
+++ b/src/gallium/drivers/r300/r300_winsys.h
@@ -48,6 +48,9 @@ struct r300_winsys {
     /* GB pipe count */
     uint32_t gb_pipes;
 
+    /* Z pipe count (rv530 only) */
+    uint32_t z_pipes;
+
     /* GART size. */
     uint32_t gart_size;
 
diff --git a/src/gallium/winsys/drm/radeon/core/radeon_r300.c b/src/gallium/winsys/drm/radeon/core/radeon_r300.c
index 3587892e00..7ea5d1fb4e 100644
--- a/src/gallium/winsys/drm/radeon/core/radeon_r300.c
+++ b/src/gallium/winsys/drm/radeon/core/radeon_r300.c
@@ -164,6 +164,16 @@ static void do_ioctls(struct r300_winsys* winsys, int fd)
     }
     winsys->gb_pipes = target;
 
+    /* get Z pipes */
+    info.request = RADEON_INFO_NUM_Z_PIPES;
+    retval = drmCommandWriteRead(fd, DRM_RADEON_INFO, &info, sizeof(info));
+    if (retval) {
+        fprintf(stderr, "%s: Failed to get GB pipe count, "
+                "error number %d\n", __FUNCTION__, retval);
+        exit(1);
+    }
+    winsys->z_pipes = target;
+
     /* Then, get PCI ID */
     info.request = RADEON_INFO_DEVICE_ID;
     retval = drmCommandWriteRead(fd, DRM_RADEON_INFO, &info, sizeof(info));