static void
nvc0_query_destroy(struct pipe_context *pipe, struct pipe_query *pq)
{
nvc0_query_allocate(nvc0_context(pipe), nvc0_query(pq), 0);
nouveau_fence_ref(NULL, &nvc0_query(pq)->fence);
FREE(nvc0_query(pq));
}
static bool
nvc0_end_query(struct pipe_context *pipe, struct pipe_query *pq)
{
struct nvc0_query *q = nvc0_query(pq);
q->funcs->end_query(nvc0_context(pipe), q);
return true;
}
static boolean
nvc0_get_query_result(struct pipe_context *pipe, struct pipe_query *pq,
boolean wait, union pipe_query_result *result)
{
struct nvc0_query *q = nvc0_query(pq);
return q->funcs->get_query_result(nvc0_context(pipe), q, wait, result);
}
static void
nvc0_render_condition(struct pipe_context *pipe,
struct pipe_query *pq,
boolean condition, uint mode)
{
struct nvc0_context *nvc0 = nvc0_context(pipe);
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
struct nvc0_query *q;
uint32_t cond;
boolean negated = FALSE;
boolean wait =
mode != PIPE_RENDER_COND_NO_WAIT &&
mode != PIPE_RENDER_COND_BY_REGION_NO_WAIT;
nvc0->cond_query = pq;
nvc0->cond_cond = condition;
nvc0->cond_mode = mode;
if (!pq) {
PUSH_SPACE(push, 1);
IMMED_NVC0(push, NVC0_3D(COND_MODE), NVC0_3D_COND_MODE_ALWAYS);
return;
}
q = nvc0_query(pq);
/* NOTE: comparison of 2 queries only works if both have completed */
switch (q->type) {
case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
cond = negated ? NVC0_3D_COND_MODE_EQUAL :
NVC0_3D_COND_MODE_NOT_EQUAL;
wait = TRUE;
break;
case PIPE_QUERY_OCCLUSION_COUNTER:
case PIPE_QUERY_OCCLUSION_PREDICATE:
if (likely(!negated)) {
if (unlikely(q->nesting))
cond = wait ? NVC0_3D_COND_MODE_NOT_EQUAL :
NVC0_3D_COND_MODE_ALWAYS;
else
cond = NVC0_3D_COND_MODE_RES_NON_ZERO;
} else {
cond = wait ? NVC0_3D_COND_MODE_EQUAL : NVC0_3D_COND_MODE_ALWAYS;
}
break;
default:
assert(!"render condition query not a predicate");
mode = NVC0_3D_COND_MODE_ALWAYS;
break;
}
if (wait)
nvc0_query_fifo_wait(push, pq);
PUSH_SPACE(push, 4);
PUSH_REFN (push, q->bo, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
BEGIN_NVC0(push, NVC0_3D(COND_ADDRESS_HIGH), 3);
PUSH_DATAh(push, q->bo->offset + q->offset);
PUSH_DATA (push, q->bo->offset + q->offset);
PUSH_DATA (push, cond);
}
void
nvc0_query_pushbuf_submit(struct nouveau_pushbuf *push,
struct pipe_query *pq, unsigned result_offset)
{
struct nvc0_query *q = nvc0_query(pq);
#define NVC0_IB_ENTRY_1_NO_PREFETCH (1 << (31 - 8))
nouveau_pushbuf_space(push, 0, 0, 1);
nouveau_pushbuf_data(push, q->bo, q->offset + result_offset, 4 |
NVC0_IB_ENTRY_1_NO_PREFETCH);
}
void
nvc0_so_target_save_offset(struct pipe_context *pipe,
struct pipe_stream_output_target *ptarg,
unsigned index, boolean *serialize)
{
struct nvc0_so_target *targ = nvc0_so_target(ptarg);
if (*serialize) {
*serialize = FALSE;
PUSH_SPACE(nvc0_context(pipe)->base.pushbuf, 1);
IMMED_NVC0(nvc0_context(pipe)->base.pushbuf, NVC0_3D(SERIALIZE), 0);
}
nvc0_query(targ->pq)->index = index;
nvc0_query_end(pipe, targ->pq);
}
void
nvc0_query_fifo_wait(struct nouveau_pushbuf *push, struct pipe_query *pq)
{
struct nvc0_query *q = nvc0_query(pq);
unsigned offset = q->offset;
if (q->type == PIPE_QUERY_SO_OVERFLOW_PREDICATE) offset += 0x20;
PUSH_SPACE(push, 5);
PUSH_REFN (push, q->bo, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
BEGIN_NVC0(push, SUBC_3D(NV84_SUBCHAN_SEMAPHORE_ADDRESS_HIGH), 4);
PUSH_DATAh(push, q->bo->offset + offset);
PUSH_DATA (push, q->bo->offset + offset);
PUSH_DATA (push, q->sequence);
PUSH_DATA (push, (1 << 12) |
NV84_SUBCHAN_SEMAPHORE_TRIGGER_ACQUIRE_EQUAL);
}
static void
nvc0_get_query_result_resource(struct pipe_context *pipe,
struct pipe_query *pq,
boolean wait,
enum pipe_query_value_type result_type,
int index,
struct pipe_resource *resource,
unsigned offset)
{
struct nvc0_query *q = nvc0_query(pq);
if (!q->funcs->get_query_result_resource) {
assert(!"Unexpected lack of get_query_result_resource");
return;
}
q->funcs->get_query_result_resource(nvc0_context(pipe), q, wait, result_type,
index, resource, offset);
}
static void
nvc0_so_target_save_offset(struct pipe_context *pipe,
struct pipe_stream_output_target *ptarg,
unsigned index, bool *serialize)
{
struct nvc0_so_target *targ = nvc0_so_target(ptarg);
if (*serialize) {
*serialize = false;
PUSH_SPACE(nvc0_context(pipe)->base.pushbuf, 1);
IMMED_NVC0(nvc0_context(pipe)->base.pushbuf, NVC0_3D(SERIALIZE), 0);
NOUVEAU_DRV_STAT(nouveau_screen(pipe->screen), gpu_serialize_count, 1);
}
nvc0_query(targ->pq)->index = index;
pipe->end_query(pipe, targ->pq);
}
static void
nvc0_render_condition(struct pipe_context *pipe,
struct pipe_query *pq,
boolean condition, enum pipe_render_cond_flag mode)
{
struct nvc0_context *nvc0 = nvc0_context(pipe);
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
struct nvc0_query *q = nvc0_query(pq);
struct nvc0_hw_query *hq = nvc0_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 = NVC0_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:
case PIPE_QUERY_SO_OVERFLOW_ANY_PREDICATE:
cond = condition ? NVC0_3D_COND_MODE_EQUAL :
NVC0_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 == NVC0_HW_QUERY_STATE_READY)
wait = true;
if (likely(!condition)) {
cond = wait ? NVC0_3D_COND_MODE_NOT_EQUAL : NVC0_3D_COND_MODE_ALWAYS;
} else {
cond = wait ? NVC0_3D_COND_MODE_EQUAL : NVC0_3D_COND_MODE_ALWAYS;
}
break;
default:
assert(!"render condition query not a predicate");
cond = NVC0_3D_COND_MODE_ALWAYS;
break;
}
}
nvc0->cond_query = pq;
nvc0->cond_cond = condition;
nvc0->cond_condmode = cond;
nvc0->cond_mode = mode;
if (!pq) {
PUSH_SPACE(push, 2);
IMMED_NVC0(push, NVC0_3D(COND_MODE), cond);
if (nvc0->screen->compute)
IMMED_NVC0(push, NVC0_CP(COND_MODE), cond);
return;
}
if (wait && hq->state != NVC0_HW_QUERY_STATE_READY)
nvc0_hw_query_fifo_wait(nvc0, q);
PUSH_SPACE(push, 10);
PUSH_REFN (push, hq->bo, NOUVEAU_BO_GART | NOUVEAU_BO_RD);
BEGIN_NVC0(push, NVC0_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_NVC0(push, NVC0_2D(COND_ADDRESS_HIGH), 2);
PUSH_DATAh(push, hq->bo->offset + hq->offset);
PUSH_DATA (push, hq->bo->offset + hq->offset);
if (nvc0->screen->compute) {
BEGIN_NVC0(push, NVC0_CP(COND_ADDRESS_HIGH), 3);
PUSH_DATAh(push, hq->bo->offset + hq->offset);
PUSH_DATA (push, hq->bo->offset + hq->offset);
PUSH_DATA (push, cond);
}
}
static void
nvc0_query_destroy(struct pipe_context *pipe, struct pipe_query *pq)
{
nvc0_query_allocate(nvc0_context(pipe), nvc0_query(pq), 0);
FREE(nvc0_query(pq));
}
void
nvc0_tfb_validate(struct nvc0_context *nvc0)
{
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
struct nvc0_transform_feedback_state *tfb;
unsigned b;
if (nvc0->gmtyprog) tfb = nvc0->gmtyprog->tfb;
else
if (nvc0->tevlprog) tfb = nvc0->tevlprog->tfb;
else
tfb = nvc0->vertprog->tfb;
IMMED_NVC0(push, NVC0_3D(TFB_ENABLE), (tfb && nvc0->num_tfbbufs) ? 1 : 0);
if (tfb && tfb != nvc0->state.tfb) {
for (b = 0; b < 4; ++b) {
if (tfb->varying_count[b]) {
unsigned n = (tfb->varying_count[b] + 3) / 4;
BEGIN_NVC0(push, NVC0_3D(TFB_STREAM(b)), 3);
PUSH_DATA (push, tfb->stream[b]);
PUSH_DATA (push, tfb->varying_count[b]);
PUSH_DATA (push, tfb->stride[b]);
BEGIN_NVC0(push, NVC0_3D(TFB_VARYING_LOCS(b, 0)), n);
PUSH_DATAp(push, tfb->varying_index[b], n);
if (nvc0->tfbbuf[b])
nvc0_so_target(nvc0->tfbbuf[b])->stride = tfb->stride[b];
} else {
IMMED_NVC0(push, NVC0_3D(TFB_VARYING_COUNT(b)), 0);
}
}
}
nvc0->state.tfb = tfb;
if (!(nvc0->dirty_3d & NVC0_NEW_3D_TFB_TARGETS))
return;
for (b = 0; b < nvc0->num_tfbbufs; ++b) {
struct nvc0_so_target *targ = nvc0_so_target(nvc0->tfbbuf[b]);
struct nv04_resource *buf;
if (!targ) {
IMMED_NVC0(push, NVC0_3D(TFB_BUFFER_ENABLE(b)), 0);
continue;
}
if (tfb)
targ->stride = tfb->stride[b];
buf = nv04_resource(targ->pipe.buffer);
BCTX_REFN(nvc0->bufctx_3d, 3D_TFB, buf, WR);
if (!(nvc0->tfbbuf_dirty & (1 << b)))
continue;
if (!targ->clean)
nvc0_hw_query_fifo_wait(nvc0, nvc0_query(targ->pq));
nouveau_pushbuf_space(push, 0, 0, 1);
BEGIN_NVC0(push, NVC0_3D(TFB_BUFFER_ENABLE(b)), 5);
PUSH_DATA (push, 1);
PUSH_DATAh(push, buf->address + targ->pipe.buffer_offset);
PUSH_DATA (push, buf->address + targ->pipe.buffer_offset);
PUSH_DATA (push, targ->pipe.buffer_size);
if (!targ->clean) {
nvc0_hw_query_pushbuf_submit(push, nvc0_query(targ->pq), 0x4);
} else {
PUSH_DATA(push, 0); /* TFB_BUFFER_OFFSET */
targ->clean = false;
}
}
for (; b < 4; ++b)
IMMED_NVC0(push, NVC0_3D(TFB_BUFFER_ENABLE(b)), 0);
}
static boolean
nvc0_begin_query(struct pipe_context *pipe, struct pipe_query *pq)
{
struct nvc0_query *q = nvc0_query(pq);
return q->funcs->begin_query(nvc0_context(pipe), q);
}
static void
nvc0_destroy_query(struct pipe_context *pipe, struct pipe_query *pq)
{
struct nvc0_query *q = nvc0_query(pq);
q->funcs->destroy_query(nvc0_context(pipe), q);
}
static void
nve4_mp_pm_query_end(struct nvc0_context *nvc0, struct nvc0_query *q)
{
struct nvc0_screen *screen = nvc0->screen;
struct pipe_context *pipe = &nvc0->base.pipe;
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
uint32_t mask;
uint32_t input[3];
const uint block[3] = { 32, 4, 1 };
const uint grid[3] = { screen->mp_count, 1, 1 };
unsigned c;
const struct nve4_mp_pm_query_cfg *cfg;
cfg = &nve4_mp_pm_queries[q->type - PIPE_QUERY_DRIVER_SPECIFIC];
if (unlikely(!screen->pm.prog)) {
struct nvc0_program *prog = CALLOC_STRUCT(nvc0_program);
prog->type = PIPE_SHADER_COMPUTE;
prog->translated = TRUE;
prog->num_gprs = 14;
prog->code = (uint32_t *)nve4_read_mp_pm_counters_code;
prog->code_size = sizeof(nve4_read_mp_pm_counters_code);
prog->parm_size = 12;
screen->pm.prog = prog;
}
/* disable all counting */
PUSH_SPACE(push, 8);
for (c = 0; c < 8; ++c)
if (screen->pm.mp_counter[c])
IMMED_NVC0(push, NVE4_COMPUTE(MP_PM_FUNC(c)), 0);
/* release counters for this query */
for (c = 0; c < 8; ++c) {
if (nvc0_query(screen->pm.mp_counter[c]) == q) {
screen->pm.num_mp_pm_active[c / 4]--;
screen->pm.mp_counter[c] = NULL;
}
}
BCTX_REFN_bo(nvc0->bufctx_cp, CP_QUERY, NOUVEAU_BO_GART | NOUVEAU_BO_WR,
q->bo);
PUSH_SPACE(push, 1);
IMMED_NVC0(push, SUBC_COMPUTE(NV50_GRAPH_SERIALIZE), 0);
pipe->bind_compute_state(pipe, screen->pm.prog);
input[0] = (q->bo->offset + q->base);
input[1] = (q->bo->offset + q->base) >> 32;
input[2] = q->sequence;
pipe->launch_grid(pipe, block, grid, 0, input);
nouveau_bufctx_reset(nvc0->bufctx_cp, NVC0_BIND_CP_QUERY);
/* re-activate other counters */
PUSH_SPACE(push, 16);
mask = 0;
for (c = 0; c < 8; ++c) {
unsigned i;
q = nvc0_query(screen->pm.mp_counter[c]);
if (!q)
continue;
cfg = &nve4_mp_pm_queries[q->type - PIPE_QUERY_DRIVER_SPECIFIC];
for (i = 0; i < cfg->num_counters; ++i) {
if (mask & (1 << q->ctr[i]))
break;
mask |= 1 << q->ctr[i];
BEGIN_NVC0(push, NVE4_COMPUTE(MP_PM_FUNC(q->ctr[i])), 1);
PUSH_DATA (push, (cfg->ctr[i].func << 4) | cfg->ctr[i].mode);
}
}
}
static void
nvc0_query_begin(struct pipe_context *pipe, struct pipe_query *pq)
{
struct nvc0_context *nvc0 = nvc0_context(pipe);
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
struct nvc0_query *q = nvc0_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->rotate) {
nvc0_query_rotate(nvc0, q);
/* 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;
}
q->sequence++;
switch (q->type) {
case PIPE_QUERY_OCCLUSION_COUNTER:
case PIPE_QUERY_OCCLUSION_PREDICATE:
q->nesting = nvc0->screen->num_occlusion_queries_active++;
if (q->nesting) {
nvc0_query_get(push, q, 0x10, 0x0100f002);
} else {
PUSH_SPACE(push, 3);
BEGIN_NVC0(push, NVC0_3D(COUNTER_RESET), 1);
PUSH_DATA (push, NVC0_3D_COUNTER_RESET_SAMPLECNT);
IMMED_NVC0(push, NVC0_3D(SAMPLECNT_ENABLE), 1);
}
break;
case PIPE_QUERY_PRIMITIVES_GENERATED:
nvc0_query_get(push, q, 0x10, 0x09005002 | (q->index << 5));
break;
case PIPE_QUERY_PRIMITIVES_EMITTED:
nvc0_query_get(push, q, 0x10, 0x05805002 | (q->index << 5));
break;
case PIPE_QUERY_SO_STATISTICS:
nvc0_query_get(push, q, 0x20, 0x05805002 | (q->index << 5));
nvc0_query_get(push, q, 0x30, 0x06805002 | (q->index << 5));
break;
case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
nvc0_query_get(push, q, 0x10, 0x03005002 | (q->index << 5));
break;
case PIPE_QUERY_TIME_ELAPSED:
nvc0_query_get(push, q, 0x10, 0x00005002);
break;
case PIPE_QUERY_PIPELINE_STATISTICS:
nvc0_query_get(push, q, 0xc0 + 0x00, 0x00801002); /* VFETCH, VERTICES */
nvc0_query_get(push, q, 0xc0 + 0x10, 0x01801002); /* VFETCH, PRIMS */
nvc0_query_get(push, q, 0xc0 + 0x20, 0x02802002); /* VP, LAUNCHES */
nvc0_query_get(push, q, 0xc0 + 0x30, 0x03806002); /* GP, LAUNCHES */
nvc0_query_get(push, q, 0xc0 + 0x40, 0x04806002); /* GP, PRIMS_OUT */
nvc0_query_get(push, q, 0xc0 + 0x50, 0x07804002); /* RAST, PRIMS_IN */
nvc0_query_get(push, q, 0xc0 + 0x60, 0x08804002); /* RAST, PRIMS_OUT */
nvc0_query_get(push, q, 0xc0 + 0x70, 0x0980a002); /* ROP, PIXELS */
nvc0_query_get(push, q, 0xc0 + 0x80, 0x0d808002); /* TCP, LAUNCHES */
nvc0_query_get(push, q, 0xc0 + 0x90, 0x0e809002); /* TEP, LAUNCHES */
break;
default:
#ifdef NOUVEAU_ENABLE_DRIVER_STATISTICS
if (q->type >= NVC0_QUERY_DRV_STAT(0) &&
q->type <= NVC0_QUERY_DRV_STAT_LAST) {
if (q->index >= 5)
q->u.value = nvc0->screen->base.stats.v[q->index];
else
q->u.value = 0;
} else
#endif
if (q->type >= NVE4_PM_QUERY(0) && q->type <= NVE4_PM_QUERY_LAST) {
nve4_mp_pm_query_begin(nvc0, q);
}
break;
}
q->state = NVC0_QUERY_STATE_ACTIVE;
}
static void
nvc0_query_end(struct pipe_context *pipe, struct pipe_query *pq)
{
struct nvc0_context *nvc0 = nvc0_context(pipe);
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
struct nvc0_query *q = nvc0_query(pq);
if (q->state != NVC0_QUERY_STATE_ACTIVE) {
/* some queries don't require 'begin' to be called (e.g. GPU_FINISHED) */
if (q->rotate)
nvc0_query_rotate(nvc0, q);
q->sequence++;
}
q->state = NVC0_QUERY_STATE_ENDED;
switch (q->type) {
case PIPE_QUERY_OCCLUSION_COUNTER:
case PIPE_QUERY_OCCLUSION_PREDICATE:
nvc0_query_get(push, q, 0, 0x0100f002);
if (--nvc0->screen->num_occlusion_queries_active == 0) {
PUSH_SPACE(push, 1);
IMMED_NVC0(push, NVC0_3D(SAMPLECNT_ENABLE), 0);
}
break;
case PIPE_QUERY_PRIMITIVES_GENERATED:
nvc0_query_get(push, q, 0, 0x09005002 | (q->index << 5));
break;
case PIPE_QUERY_PRIMITIVES_EMITTED:
nvc0_query_get(push, q, 0, 0x05805002 | (q->index << 5));
break;
case PIPE_QUERY_SO_STATISTICS:
nvc0_query_get(push, q, 0x00, 0x05805002 | (q->index << 5));
nvc0_query_get(push, q, 0x10, 0x06805002 | (q->index << 5));
break;
case PIPE_QUERY_SO_OVERFLOW_PREDICATE:
/* TODO: How do we sum over all streams for render condition ? */
/* PRIMS_DROPPED doesn't write sequence, use a ZERO query to sync on */
nvc0_query_get(push, q, 0x00, 0x03005002 | (q->index << 5));
nvc0_query_get(push, q, 0x20, 0x00005002);
break;
case PIPE_QUERY_TIMESTAMP:
case PIPE_QUERY_TIME_ELAPSED:
nvc0_query_get(push, q, 0, 0x00005002);
break;
case PIPE_QUERY_GPU_FINISHED:
nvc0_query_get(push, q, 0, 0x1000f010);
break;
case PIPE_QUERY_PIPELINE_STATISTICS:
nvc0_query_get(push, q, 0x00, 0x00801002); /* VFETCH, VERTICES */
nvc0_query_get(push, q, 0x10, 0x01801002); /* VFETCH, PRIMS */
nvc0_query_get(push, q, 0x20, 0x02802002); /* VP, LAUNCHES */
nvc0_query_get(push, q, 0x30, 0x03806002); /* GP, LAUNCHES */
nvc0_query_get(push, q, 0x40, 0x04806002); /* GP, PRIMS_OUT */
nvc0_query_get(push, q, 0x50, 0x07804002); /* RAST, PRIMS_IN */
nvc0_query_get(push, q, 0x60, 0x08804002); /* RAST, PRIMS_OUT */
nvc0_query_get(push, q, 0x70, 0x0980a002); /* ROP, PIXELS */
nvc0_query_get(push, q, 0x80, 0x0d808002); /* TCP, LAUNCHES */
nvc0_query_get(push, q, 0x90, 0x0e809002); /* TEP, LAUNCHES */
break;
case NVC0_QUERY_TFB_BUFFER_OFFSET:
/* indexed by TFB buffer instead of by vertex stream */
nvc0_query_get(push, q, 0x00, 0x0d005002 | (q->index << 5));
break;
default:
#ifdef NOUVEAU_ENABLE_DRIVER_STATISTICS
if (q->type >= NVC0_QUERY_DRV_STAT(0) &&
q->type <= NVC0_QUERY_DRV_STAT_LAST) {
q->u.value = nvc0->screen->base.stats.v[q->index] - q->u.value;
return;
} else
#endif
if (q->type >= NVE4_PM_QUERY(0) && q->type <= NVE4_PM_QUERY_LAST)
nve4_mp_pm_query_end(nvc0, q);
break;
}
if (q->is64bit)
nouveau_fence_ref(nvc0->screen->base.fence.current, &q->fence);
}
static boolean
nvc0_query_result(struct pipe_context *pipe, struct pipe_query *pq,
boolean wait, union pipe_query_result *result)
{
struct nvc0_context *nvc0 = nvc0_context(pipe);
struct nvc0_query *q = nvc0_query(pq);
uint64_t *res64 = (uint64_t*)result;
uint32_t *res32 = (uint32_t*)result;
boolean *res8 = (boolean*)result;
uint64_t *data64 = (uint64_t *)q->data;
unsigned i;
if (q->state != NVC0_QUERY_STATE_READY)
nvc0_query_update(nvc0->screen->base.client, q);
if (q->state != NVC0_QUERY_STATE_READY) {
if (!wait) {
if (q->state != NVC0_QUERY_STATE_FLUSHED) {
q->state = NVC0_QUERY_STATE_FLUSHED;
/* flush for silly apps that spin on GL_QUERY_RESULT_AVAILABLE */
PUSH_KICK(nvc0->base.pushbuf);
}
return FALSE;
}
if (nouveau_bo_wait(q->bo, NOUVEAU_BO_RD, nvc0->screen->base.client))
return FALSE;
}
q->state = NVC0_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] = q->data[1] - q->data[5];
break;
case PIPE_QUERY_OCCLUSION_PREDICATE:
res8[0] = q->data[1] != q->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_SO_OVERFLOW_PREDICATE:
res8[0] = data64[0] != data64[2];
break;
case PIPE_QUERY_TIMESTAMP:
res64[0] = data64[1];
break;
case PIPE_QUERY_TIMESTAMP_DISJOINT: /* u32 sequence, u32 0, u64 time */
res64[0] = 1000000000;
res8[8] = (data64[1] == data64[3]) ? FALSE : TRUE;
break;
case PIPE_QUERY_TIME_ELAPSED:
res64[0] = data64[1] - data64[3];
break;
case PIPE_QUERY_PIPELINE_STATISTICS:
for (i = 0; i < 10; ++i)
res64[i] = data64[i * 2] - data64[24 + i * 2];
break;
case NVC0_QUERY_TFB_BUFFER_OFFSET:
res32[0] = q->data[1];
break;
default:
return FALSE;
}
return TRUE;
}
static boolean
nvc0_query_result(struct pipe_context *pipe, struct pipe_query *pq,
boolean wait, union pipe_query_result *result)
{
struct nvc0_context *nvc0 = nvc0_context(pipe);
struct nvc0_query *q = nvc0_query(pq);
uint64_t *res64 = (uint64_t*)result;
uint32_t *res32 = (uint32_t*)result;
boolean *res8 = (boolean*)result;
uint64_t *data64 = (uint64_t *)q->data;
unsigned i;
#ifdef NOUVEAU_ENABLE_DRIVER_STATISTICS
if (q->type >= NVC0_QUERY_DRV_STAT(0) &&
q->type <= NVC0_QUERY_DRV_STAT_LAST) {
res64[0] = q->u.value;
return TRUE;
} else
#endif
if (q->type >= NVE4_PM_QUERY(0) && q->type <= NVE4_PM_QUERY_LAST) {
return nve4_mp_pm_query_result(nvc0, q, result, wait);
}
if (q->state != NVC0_QUERY_STATE_READY)
nvc0_query_update(nvc0->screen->base.client, q);
if (q->state != NVC0_QUERY_STATE_READY) {
if (!wait) {
if (q->state != NVC0_QUERY_STATE_FLUSHED) {
q->state = NVC0_QUERY_STATE_FLUSHED;
/* flush for silly apps that spin on GL_QUERY_RESULT_AVAILABLE */
PUSH_KICK(nvc0->base.pushbuf);
}
return FALSE;
}
if (nouveau_bo_wait(q->bo, NOUVEAU_BO_RD, nvc0->screen->base.client))
return FALSE;
NOUVEAU_DRV_STAT(&nvc0->screen->base, query_sync_count, 1);
}
q->state = NVC0_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] = q->data[1] - q->data[5];
break;
case PIPE_QUERY_OCCLUSION_PREDICATE:
res8[0] = q->data[1] != q->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_SO_OVERFLOW_PREDICATE:
res8[0] = data64[0] != data64[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 PIPE_QUERY_PIPELINE_STATISTICS:
for (i = 0; i < 10; ++i)
res64[i] = data64[i * 2] - data64[24 + i * 2];
break;
case NVC0_QUERY_TFB_BUFFER_OFFSET:
res32[0] = q->data[1];
break;
default:
assert(0); /* can't happen, we don't create queries with invalid type */
return FALSE;
}
return TRUE;
}
