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;
}
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 inline void
nv50_hw_query_update(struct nv50_query *q)
{
struct nv50_hw_query *hq = nv50_hw_query(q);
if (hq->is64bit) {
if (nouveau_fence_signalled(hq->fence))
hq->state = NV50_HW_QUERY_STATE_READY;
} else {
if (hq->data[0] == hq->sequence)
hq->state = NV50_HW_QUERY_STATE_READY;
}
}
void
nv84_hw_query_fifo_wait(struct nouveau_pushbuf *push, struct nv50_query *q)
{
struct nv50_hw_query *hq = nv50_hw_query(q);
unsigned offset = hq->offset;
PUSH_SPACE(push, 5);
PUSH_REFN (push, hq->bo, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
BEGIN_NV04(push, SUBC_3D(NV84_SUBCHAN_SEMAPHORE_ADDRESS_HIGH), 4);
PUSH_DATAh(push, hq->bo->offset + offset);
PUSH_DATA (push, hq->bo->offset + offset);
PUSH_DATA (push, hq->sequence);
PUSH_DATA (push, NV84_SUBCHAN_SEMAPHORE_TRIGGER_ACQUIRE_EQUAL);
}
void
nv50_hw_query_pushbuf_submit(struct nouveau_pushbuf *push, uint16_t method,
struct nv50_query *q, unsigned result_offset)
{
struct nv50_hw_query *hq = nv50_hw_query(q);
nv50_hw_query_update(q);
if (hq->state != NV50_HW_QUERY_STATE_READY)
nouveau_bo_wait(hq->bo, NOUVEAU_BO_RD, push->client);
hq->state = NV50_HW_QUERY_STATE_READY;
BEGIN_NV04(push, SUBC_3D(method), 1);
PUSH_DATA (push, hq->data[result_offset / 4]);
}
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);
}
static void
nv50_hw_query_get(struct nouveau_pushbuf *push, struct nv50_query *q,
unsigned offset, uint32_t get)
{
struct nv50_hw_query *hq = nv50_hw_query(q);
offset += hq->offset;
PUSH_SPACE(push, 5);
PUSH_REFN (push, hq->bo, NOUVEAU_BO_GART | NOUVEAU_BO_WR);
BEGIN_NV04(push, NV50_3D(QUERY_ADDRESS_HIGH), 4);
PUSH_DATAh(push, hq->bo->offset + offset);
PUSH_DATA (push, hq->bo->offset + offset);
PUSH_DATA (push, hq->sequence);
PUSH_DATA (push, get);
}
static boolean
nv50_hw_get_query_result(struct nv50_context *nv50, struct nv50_query *q,
boolean wait, union pipe_query_result *result)
{
struct nv50_hw_query *hq = nv50_hw_query(q);
uint64_t *res64 = (uint64_t *)result;
uint32_t *res32 = (uint32_t *)result;
uint8_t *res8 = (uint8_t *)result;
uint64_t *data64 = (uint64_t *)hq->data;
int i;
if (hq->state != NV50_HW_QUERY_STATE_READY)
nv50_hw_query_update(q);
if (hq->state != NV50_HW_QUERY_STATE_READY) {
if (!wait) {
/* for broken apps that spin on GL_QUERY_RESULT_AVAILABLE */
if (hq->state != NV50_HW_QUERY_STATE_FLUSHED) {
hq->state = NV50_HW_QUERY_STATE_FLUSHED;
PUSH_KICK(nv50->base.pushbuf);
}
return false;
}
if (nouveau_bo_wait(hq->bo, NOUVEAU_BO_RD, nv50->screen->base.client))
return false;
}
hq->state = NV50_HW_QUERY_STATE_READY;
switch (q->type) {
case PIPE_QUERY_GPU_FINISHED:
res8[0] = true;
break;
case PIPE_QUERY_OCCLUSION_COUNTER: /* u32 sequence, u32 count, u64 time */
res64[0] = hq->data[1] - hq->data[5];
break;
case PIPE_QUERY_PRIMITIVES_GENERATED: /* u64 count, u64 time */
case PIPE_QUERY_PRIMITIVES_EMITTED: /* u64 count, u64 time */
res64[0] = data64[0] - data64[2];
break;
case PIPE_QUERY_SO_STATISTICS:
res64[0] = data64[0] - data64[4];
res64[1] = data64[2] - data64[6];
break;
case PIPE_QUERY_PIPELINE_STATISTICS:
for (i = 0; i < 8; ++i)
res64[i] = data64[i * 2] - data64[16 + i * 2];
break;
case PIPE_QUERY_TIMESTAMP:
res64[0] = data64[1];
break;
case PIPE_QUERY_TIMESTAMP_DISJOINT:
res64[0] = 1000000000;
res8[8] = false;
break;
case PIPE_QUERY_TIME_ELAPSED:
res64[0] = data64[1] - data64[3];
break;
case NVA0_HW_QUERY_STREAM_OUTPUT_BUFFER_OFFSET:
res32[0] = hq->data[1];
break;
default:
assert(0);
return false;
}
return true;
}
static void
nv50_hw_end_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);
hq->state = NV50_HW_QUERY_STATE_ENDED;
switch (q->type) {
case PIPE_QUERY_OCCLUSION_COUNTER:
nv50_hw_query_get(push, q, 0, 0x0100f002);
if (--nv50->screen->num_occlusion_queries_active == 0) {
PUSH_SPACE(push, 2);
BEGIN_NV04(push, NV50_3D(SAMPLECNT_ENABLE), 1);
PUSH_DATA (push, 0);
}
break;
case PIPE_QUERY_PRIMITIVES_GENERATED:
nv50_hw_query_get(push, q, 0, 0x06805002);
break;
case PIPE_QUERY_PRIMITIVES_EMITTED:
nv50_hw_query_get(push, q, 0, 0x05805002);
break;
case PIPE_QUERY_SO_STATISTICS:
nv50_hw_query_get(push, q, 0x00, 0x05805002);
nv50_hw_query_get(push, q, 0x10, 0x06805002);
break;
case PIPE_QUERY_PIPELINE_STATISTICS:
nv50_hw_query_get(push, q, 0x00, 0x00801002); /* VFETCH, VERTICES */
nv50_hw_query_get(push, q, 0x10, 0x01801002); /* VFETCH, PRIMS */
nv50_hw_query_get(push, q, 0x20, 0x02802002); /* VP, LAUNCHES */
nv50_hw_query_get(push, q, 0x30, 0x03806002); /* GP, LAUNCHES */
nv50_hw_query_get(push, q, 0x40, 0x04806002); /* GP, PRIMS_OUT */
nv50_hw_query_get(push, q, 0x50, 0x07804002); /* RAST, PRIMS_IN */
nv50_hw_query_get(push, q, 0x60, 0x08804002); /* RAST, PRIMS_OUT */
nv50_hw_query_get(push, q, 0x70, 0x0980a002); /* ROP, PIXELS */
break;
case PIPE_QUERY_TIMESTAMP:
hq->sequence++;
/* fall through */
case PIPE_QUERY_TIME_ELAPSED:
nv50_hw_query_get(push, q, 0, 0x00005002);
break;
case PIPE_QUERY_GPU_FINISHED:
hq->sequence++;
nv50_hw_query_get(push, q, 0, 0x1000f010);
break;
case NVA0_HW_QUERY_STREAM_OUTPUT_BUFFER_OFFSET:
hq->sequence++;
nv50_hw_query_get(push, q, 0, 0x0d005002 | (q->index << 5));
break;
case PIPE_QUERY_TIMESTAMP_DISJOINT:
/* This query is not issued on GPU because disjoint is forced to false */
hq->state = NV50_HW_QUERY_STATE_READY;
break;
default:
assert(0);
break;
}
if (hq->is64bit)
nouveau_fence_ref(nv50->screen->base.fence.current, &hq->fence);
}
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;
}
static void
nv50_render_condition(struct pipe_context *pipe,
struct pipe_query *pq,
boolean condition, enum pipe_render_cond_flag mode)
{
struct nv50_context *nv50 = nv50_context(pipe);
struct nouveau_pushbuf *push = nv50->base.pushbuf;
struct nv50_query *q = nv50_query(pq);
struct nv50_hw_query *hq = nv50_hw_query(q);
uint32_t cond;
bool wait =
mode != PIPE_RENDER_COND_NO_WAIT &&
mode != PIPE_RENDER_COND_BY_REGION_NO_WAIT;
if (!pq) {
cond = NV50_3D_COND_MODE_ALWAYS;
}
else {
/* NOTE: comparison of 2 queries only works if both have completed */
switch (q->type) {
case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
cond = condition ? NV50_3D_COND_MODE_EQUAL :
NV50_3D_COND_MODE_NOT_EQUAL;
wait = true;
break;
case PIPE_QUERY_OCCLUSION_COUNTER:
case PIPE_QUERY_OCCLUSION_PREDICATE:
case PIPE_QUERY_OCCLUSION_PREDICATE_CONSERVATIVE:
if (hq->state == NV50_HW_QUERY_STATE_READY)
wait = true;
if (likely(!condition)) {
cond = wait ? NV50_3D_COND_MODE_NOT_EQUAL : NV50_3D_COND_MODE_ALWAYS;
} else {
cond = wait ? NV50_3D_COND_MODE_EQUAL : NV50_3D_COND_MODE_ALWAYS;
}
break;
default:
assert(!"render condition query not a predicate");
cond = NV50_3D_COND_MODE_ALWAYS;
break;
}
}
nv50->cond_query = pq;
nv50->cond_cond = condition;
nv50->cond_condmode = cond;
nv50->cond_mode = mode;
if (!pq) {
PUSH_SPACE(push, 2);
BEGIN_NV04(push, NV50_3D(COND_MODE), 1);
PUSH_DATA (push, cond);
return;
}
PUSH_SPACE(push, 9);
if (wait && hq->state != NV50_HW_QUERY_STATE_READY) {
BEGIN_NV04(push, SUBC_3D(NV50_GRAPH_SERIALIZE), 1);
PUSH_DATA (push, 0);
}
PUSH_REFN (push, hq->bo, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
BEGIN_NV04(push, NV50_3D(COND_ADDRESS_HIGH), 3);
PUSH_DATAh(push, hq->bo->offset + hq->offset);
PUSH_DATA (push, hq->bo->offset + hq->offset);
PUSH_DATA (push, cond);
BEGIN_NV04(push, NV50_2D(COND_ADDRESS_HIGH), 2);
PUSH_DATAh(push, hq->bo->offset + hq->offset);
PUSH_DATA (push, hq->bo->offset + hq->offset);
}
