static struct pipe_query *
nv50_query_create(struct pipe_context *pipe, unsigned type, unsigned index)
{
struct nv50_context *nv50 = nv50_context(pipe);
struct nv50_query *q;
q = CALLOC_STRUCT(nv50_query);
if (!q)
return NULL;
if (!nv50_query_allocate(nv50, q, NV50_QUERY_ALLOC_SPACE)) {
FREE(q);
return NULL;
}
q->is64bit = (type == PIPE_QUERY_PRIMITIVES_GENERATED ||
type == PIPE_QUERY_PRIMITIVES_EMITTED ||
type == PIPE_QUERY_SO_STATISTICS);
q->type = type;
if (q->type == PIPE_QUERY_OCCLUSION_COUNTER) {
q->offset -= 32;
q->data -= 32 / sizeof(*q->data); /* we advance before query_begin ! */
}
return (struct pipe_query *)q;
}
static boolean
nv50_query_allocate(struct nv50_context *nv50, struct nv50_query *q, int size)
{
struct nv50_screen *screen = nv50->screen;
int ret;
if (q->bo) {
nouveau_bo_ref(NULL, &q->bo);
if (q->mm) {
if (q->ready)
nouveau_mm_free(q->mm);
else
nouveau_fence_work(screen->base.fence.current, nouveau_mm_free_work,
q->mm);
}
}
if (size) {
q->mm = nouveau_mm_allocate(screen->base.mm_GART, size, &q->bo, &q->base);
if (!q->bo)
return FALSE;
q->offset = q->base;
ret = nouveau_bo_map(q->bo, 0, screen->base.client);
if (ret) {
nv50_query_allocate(nv50, q, 0);
return FALSE;
}
q->data = (uint32_t *)((uint8_t *)q->bo->map + q->base);
}
return TRUE;
}
static void
nv50_query_destroy(struct pipe_context *pipe, struct pipe_query *pq)
{
nv50_query_allocate(nv50_context(pipe), nv50_query(pq), 0);
nouveau_fence_ref(NULL, &nv50_query(pq)->fence);
FREE(nv50_query(pq));
}
static void
nv50_query_begin(struct pipe_context *pipe, struct pipe_query *pq)
{
struct nv50_context *nv50 = nv50_context(pipe);
struct nouveau_pushbuf *push = nv50->base.pushbuf;
struct nv50_query *q = nv50_query(pq);
/* For occlusion queries we have to change the storage, because a previous
* query might set the initial render conition to FALSE even *after* we re-
* initialized it to TRUE.
*/
if (q->type == PIPE_QUERY_OCCLUSION_COUNTER) {
q->offset += 16;
q->data += 16 / sizeof(*q->data);
if (q->offset - q->base == NV50_QUERY_ALLOC_SPACE)
nv50_query_allocate(nv50, q, NV50_QUERY_ALLOC_SPACE);
/* XXX: can we do this with the GPU, and sync with respect to a previous
* query ?
*/
q->data[1] = 1; /* initial render condition = TRUE */
}
if (!q->is64bit)
q->data[0] = q->sequence++; /* the previously used one */
switch (q->type) {
case PIPE_QUERY_OCCLUSION_COUNTER:
PUSH_SPACE(push, 4);
BEGIN_NV04(push, NV50_3D(COUNTER_RESET), 1);
PUSH_DATA (push, NV50_3D_COUNTER_RESET_SAMPLECNT);
BEGIN_NV04(push, NV50_3D(SAMPLECNT_ENABLE), 1);
PUSH_DATA (push, 1);
break;
case PIPE_QUERY_PRIMITIVES_GENERATED: /* store before & after instead ? */
PUSH_SPACE(push, 2);
BEGIN_NV04(push, NV50_3D(COUNTER_RESET), 1);
PUSH_DATA (push, NV50_3D_COUNTER_RESET_GENERATED_PRIMITIVES);
break;
case PIPE_QUERY_PRIMITIVES_EMITTED:
PUSH_SPACE(push, 2);
BEGIN_NV04(push, NV50_3D(COUNTER_RESET), 1);
PUSH_DATA (push, NV50_3D_COUNTER_RESET_TRANSFORM_FEEDBACK);
break;
case PIPE_QUERY_SO_STATISTICS:
PUSH_SPACE(push, 3);
BEGIN_NI04(push, NV50_3D(COUNTER_RESET), 2);
PUSH_DATA (push, NV50_3D_COUNTER_RESET_TRANSFORM_FEEDBACK);
PUSH_DATA (push, NV50_3D_COUNTER_RESET_GENERATED_PRIMITIVES);
break;
case PIPE_QUERY_TIMESTAMP_DISJOINT:
case PIPE_QUERY_TIME_ELAPSED:
nv50_query_get(push, q, 0x10, 0x00005002);
break;
default:
break;
}
q->ready = FALSE;
}
static void
nv50_query_destroy(struct pipe_context *pipe, struct pipe_query *pq)
{
nv50_query_allocate(nv50_context(pipe), nv50_query(pq), 0);
FREE(nv50_query(pq));
}
static boolean
nv50_query_begin(struct pipe_context *pipe, struct pipe_query *pq)
{
struct nv50_context *nv50 = nv50_context(pipe);
struct nouveau_pushbuf *push = nv50->base.pushbuf;
struct nv50_query *q = nv50_query(pq);
/* For occlusion queries we have to change the storage, because a previous
* query might set the initial render conition to FALSE even *after* we re-
* initialized it to TRUE.
*/
if (q->type == PIPE_QUERY_OCCLUSION_COUNTER) {
q->offset += 32;
q->data += 32 / sizeof(*q->data);
if (q->offset - q->base == NV50_QUERY_ALLOC_SPACE)
nv50_query_allocate(nv50, q, NV50_QUERY_ALLOC_SPACE);
/* XXX: can we do this with the GPU, and sync with respect to a previous
* query ?
*/
q->data[0] = q->sequence; /* initialize sequence */
q->data[1] = 1; /* initial render condition = TRUE */
q->data[4] = q->sequence + 1; /* for comparison COND_MODE */
q->data[5] = 0;
}
if (!q->is64bit)
q->data[0] = q->sequence++; /* the previously used one */
switch (q->type) {
case PIPE_QUERY_OCCLUSION_COUNTER:
PUSH_SPACE(push, 4);
BEGIN_NV04(push, NV50_3D(COUNTER_RESET), 1);
PUSH_DATA (push, NV50_3D_COUNTER_RESET_SAMPLECNT);
BEGIN_NV04(push, NV50_3D(SAMPLECNT_ENABLE), 1);
PUSH_DATA (push, 1);
break;
case PIPE_QUERY_PRIMITIVES_GENERATED:
nv50_query_get(push, q, 0x10, 0x06805002);
break;
case PIPE_QUERY_PRIMITIVES_EMITTED:
nv50_query_get(push, q, 0x10, 0x05805002);
break;
case PIPE_QUERY_SO_STATISTICS:
nv50_query_get(push, q, 0x20, 0x05805002);
nv50_query_get(push, q, 0x30, 0x06805002);
break;
case PIPE_QUERY_PIPELINE_STATISTICS:
nv50_query_get(push, q, 0x80, 0x00801002); /* VFETCH, VERTICES */
nv50_query_get(push, q, 0x90, 0x01801002); /* VFETCH, PRIMS */
nv50_query_get(push, q, 0xa0, 0x02802002); /* VP, LAUNCHES */
nv50_query_get(push, q, 0xb0, 0x03806002); /* GP, LAUNCHES */
nv50_query_get(push, q, 0xc0, 0x04806002); /* GP, PRIMS_OUT */
nv50_query_get(push, q, 0xd0, 0x07804002); /* RAST, PRIMS_IN */
nv50_query_get(push, q, 0xe0, 0x08804002); /* RAST, PRIMS_OUT */
nv50_query_get(push, q, 0xf0, 0x0980a002); /* ROP, PIXELS */
break;
case PIPE_QUERY_TIME_ELAPSED:
nv50_query_get(push, q, 0x10, 0x00005002);
break;
default:
break;
}
q->ready = FALSE;
return true;
}
