static bool
nv50_hw_query_allocate(struct nv50_context *nv50, struct nv50_query *q,
int size)
{
struct nv50_screen *screen = nv50->screen;
struct nv50_hw_query *hq = nv50_hw_query(q);
int ret;
if (hq->bo) {
nouveau_bo_ref(NULL, &hq->bo);
if (hq->mm) {
if (hq->state == NV50_HW_QUERY_STATE_READY)
nouveau_mm_free(hq->mm);
else
nouveau_fence_work(screen->base.fence.current,
nouveau_mm_free_work, hq->mm);
}
}
if (size) {
hq->mm = nouveau_mm_allocate(screen->base.mm_GART, size,
&hq->bo, &hq->base_offset);
if (!hq->bo)
return false;
hq->offset = hq->base_offset;
ret = nouveau_bo_map(hq->bo, 0, screen->base.client);
if (ret) {
nv50_hw_query_allocate(nv50, q, 0);
return false;
}
hq->data = (uint32_t *)((uint8_t *)hq->bo->map + hq->base_offset);
}
return true;
}
struct nv50_hw_query *
nv50_hw_sm_create_query(struct nv50_context *nv50, unsigned type)
{
struct nv50_hw_sm_query *hsq;
struct nv50_hw_query *hq;
unsigned space;
if (type < NV50_HW_SM_QUERY(0) || type > NV50_HW_SM_QUERY_LAST)
return NULL;
hsq = CALLOC_STRUCT(nv50_hw_sm_query);
if (!hsq)
return NULL;
hq = &hsq->base;
hq->funcs = &hw_sm_query_funcs;
hq->base.type = type;
/*
* for each MP:
* [00] = MP.C0
* [04] = MP.C1
* [08] = MP.C2
* [0c] = MP.C3
* [10] = MP.sequence
*/
space = (4 + 1) * nv50->screen->MPsInTP * sizeof(uint32_t);
if (!nv50_hw_query_allocate(nv50, &hq->base, space)) {
FREE(hq);
return NULL;
}
return hq;
}
static void
nv50_hw_destroy_query(struct nv50_context *nv50, struct nv50_query *q)
{
struct nv50_hw_query *hq = nv50_hw_query(q);
nv50_hw_query_allocate(nv50, q, 0);
nouveau_fence_ref(NULL, &hq->fence);
FREE(hq);
}
static void
nv50_hw_destroy_query(struct nv50_context *nv50, struct nv50_query *q)
{
struct nv50_hw_query *hq = nv50_hw_query(q);
if (hq->funcs && hq->funcs->destroy_query) {
hq->funcs->destroy_query(nv50, hq);
return;
}
nv50_hw_query_allocate(nv50, q, 0);
nouveau_fence_ref(NULL, &hq->fence);
FREE(hq);
}
struct nv50_query *
nv50_hw_create_query(struct nv50_context *nv50, unsigned type, unsigned index)
{
struct nv50_hw_query *hq;
struct nv50_query *q;
hq = CALLOC_STRUCT(nv50_hw_query);
if (!hq)
return NULL;
q = &hq->base;
q->funcs = &hw_query_funcs;
q->type = type;
switch (q->type) {
case PIPE_QUERY_OCCLUSION_COUNTER:
hq->rotate = 32;
break;
case PIPE_QUERY_PRIMITIVES_GENERATED:
case PIPE_QUERY_PRIMITIVES_EMITTED:
case PIPE_QUERY_SO_STATISTICS:
case PIPE_QUERY_PIPELINE_STATISTICS:
hq->is64bit = true;
break;
case PIPE_QUERY_TIME_ELAPSED:
case PIPE_QUERY_TIMESTAMP:
case PIPE_QUERY_TIMESTAMP_DISJOINT:
case PIPE_QUERY_GPU_FINISHED:
case NVA0_HW_QUERY_STREAM_OUTPUT_BUFFER_OFFSET:
break;
default:
debug_printf("invalid query type: %u\n", type);
FREE(q);
return NULL;
}
if (!nv50_hw_query_allocate(nv50, q, NV50_HW_QUERY_ALLOC_SPACE)) {
FREE(hq);
return NULL;
}
if (hq->rotate) {
/* we advance before query_begin ! */
hq->offset -= hq->rotate;
hq->data -= hq->rotate / sizeof(*hq->data);
}
return q;
}
static boolean
nv50_hw_begin_query(struct nv50_context *nv50, struct nv50_query *q)
{
struct nouveau_pushbuf *push = nv50->base.pushbuf;
struct nv50_hw_query *hq = nv50_hw_query(q);
/* For occlusion queries we have to change the storage, because a previous
* query might set the initial render condition to false even *after* we re-
* initialized it to true.
*/
if (hq->rotate) {
hq->offset += hq->rotate;
hq->data += hq->rotate / sizeof(*hq->data);
if (hq->offset - hq->base_offset == NV50_HW_QUERY_ALLOC_SPACE)
nv50_hw_query_allocate(nv50, q, NV50_HW_QUERY_ALLOC_SPACE);
/* XXX: can we do this with the GPU, and sync with respect to a previous
* query ?
*/
hq->data[0] = hq->sequence; /* initialize sequence */
hq->data[1] = 1; /* initial render condition = true */
hq->data[4] = hq->sequence + 1; /* for comparison COND_MODE */
hq->data[5] = 0;
}
if (!hq->is64bit)
hq->data[0] = hq->sequence++; /* the previously used one */
switch (q->type) {
case PIPE_QUERY_OCCLUSION_COUNTER:
hq->nesting = nv50->screen->num_occlusion_queries_active++;
if (hq->nesting) {
nv50_hw_query_get(push, q, 0x10, 0x0100f002);
} else {
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_hw_query_get(push, q, 0x10, 0x06805002);
break;
case PIPE_QUERY_PRIMITIVES_EMITTED:
nv50_hw_query_get(push, q, 0x10, 0x05805002);
break;
case PIPE_QUERY_SO_STATISTICS:
nv50_hw_query_get(push, q, 0x20, 0x05805002);
nv50_hw_query_get(push, q, 0x30, 0x06805002);
break;
case PIPE_QUERY_PIPELINE_STATISTICS:
nv50_hw_query_get(push, q, 0x80, 0x00801002); /* VFETCH, VERTICES */
nv50_hw_query_get(push, q, 0x90, 0x01801002); /* VFETCH, PRIMS */
nv50_hw_query_get(push, q, 0xa0, 0x02802002); /* VP, LAUNCHES */
nv50_hw_query_get(push, q, 0xb0, 0x03806002); /* GP, LAUNCHES */
nv50_hw_query_get(push, q, 0xc0, 0x04806002); /* GP, PRIMS_OUT */
nv50_hw_query_get(push, q, 0xd0, 0x07804002); /* RAST, PRIMS_IN */
nv50_hw_query_get(push, q, 0xe0, 0x08804002); /* RAST, PRIMS_OUT */
nv50_hw_query_get(push, q, 0xf0, 0x0980a002); /* ROP, PIXELS */
break;
case PIPE_QUERY_TIME_ELAPSED:
nv50_hw_query_get(push, q, 0x10, 0x00005002);
break;
default:
assert(0);
return false;
}
hq->state = NV50_HW_QUERY_STATE_ACTIVE;
return true;
}