void nvc0_validate_textures(struct nvc0_context *nvc0)
{
boolean need_flush;
if (nvc0->screen->base.class_3d >= NVE4_3D_CLASS) {
need_flush = nve4_validate_tic(nvc0, 0);
need_flush |= nve4_validate_tic(nvc0, 3);
need_flush |= nve4_validate_tic(nvc0, 4);
} else {
need_flush = nvc0_validate_tic(nvc0, 0);
need_flush |= nvc0_validate_tic(nvc0, 3);
need_flush |= nvc0_validate_tic(nvc0, 4);
}
if (need_flush) {
BEGIN_NVC0(nvc0->base.pushbuf, NVC0_3D(TIC_FLUSH), 1);
PUSH_DATA (nvc0->base.pushbuf, 0);
}
}
static boolean
nvc0_validate_tsc(struct nvc0_context *nvc0, int s)
{
uint32_t commands[16];
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
unsigned i;
unsigned n = 0;
boolean need_flush = FALSE;
for (i = 0; i < nvc0->num_samplers[s]; ++i) {
struct nv50_tsc_entry *tsc = nv50_tsc_entry(nvc0->samplers[s][i]);
if (!(nvc0->samplers_dirty[s] & (1 << i)))
continue;
if (!tsc) {
commands[n++] = (i << 4) | 0;
continue;
}
if (tsc->id < 0) {
tsc->id = nvc0_screen_tsc_alloc(nvc0->screen, tsc);
nvc0_m2mf_push_linear(&nvc0->base, nvc0->screen->txc,
65536 + tsc->id * 32, NOUVEAU_BO_VRAM,
32, tsc->tsc);
need_flush = TRUE;
}
nvc0->screen->tsc.lock[tsc->id / 32] |= 1 << (tsc->id % 32);
commands[n++] = (tsc->id << 12) | (i << 4) | 1;
}
for (; i < nvc0->state.num_samplers[s]; ++i)
commands[n++] = (i << 4) | 0;
nvc0->state.num_samplers[s] = nvc0->num_samplers[s];
if (n) {
BEGIN_NIC0(push, NVC0_3D(BIND_TSC(s)), n);
PUSH_DATAp(push, commands, n);
}
nvc0->samplers_dirty[s] = 0;
return need_flush;
}
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);
NOUVEAU_DRV_STAT(nouveau_screen(pipe->screen), gpu_serialize_count, 1);
}
nvc0_query(targ->pq)->index = index;
nvc0_query_end(pipe, targ->pq);
}
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 = 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:
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:
if (likely(!condition)) {
if (unlikely(hq->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");
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, 1);
IMMED_NVC0(push, NVC0_3D(COND_MODE), cond);
return;
}
if (wait)
nvc0_hw_query_fifo_wait(nvc0, q);
PUSH_SPACE(push, 7);
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);
}
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, 0);
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 & NVC0_NEW_TFB_TARGETS))
return;
nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_TFB);
for (b = 0; b < nvc0->num_tfbbufs; ++b) {
struct nvc0_so_target *targ = nvc0_so_target(nvc0->tfbbuf[b]);
struct nv04_resource *buf = nv04_resource(targ->pipe.buffer);
if (tfb)
targ->stride = tfb->stride[b];
if (!(nvc0->tfbbuf_dirty & (1 << b)))
continue;
if (!targ->clean)
nvc0_query_fifo_wait(push, targ->pq);
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_query_pushbuf_submit(push, targ->pq, 0x4);
} else {
PUSH_DATA(push, 0); /* TFB_BUFFER_OFFSET */
targ->clean = FALSE;
}
BCTX_REFN(nvc0->bufctx_3d, TFB, buf, WR);
}
for (; b < 4; ++b)
IMMED_NVC0(push, NVC0_3D(TFB_BUFFER_ENABLE(b)), 0);
}
static boolean
nvc0_validate_tic(struct nvc0_context *nvc0, int s)
{
uint32_t commands[32];
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
struct nouveau_bo *txc = nvc0->screen->txc;
unsigned i;
unsigned n = 0;
boolean need_flush = FALSE;
for (i = 0; i < nvc0->num_textures[s]; ++i) {
struct nv50_tic_entry *tic = nv50_tic_entry(nvc0->textures[s][i]);
struct nv04_resource *res;
const boolean dirty = !!(nvc0->textures_dirty[s] & (1 << i));
if (!tic) {
if (dirty)
commands[n++] = (i << 1) | 0;
continue;
}
res = nv04_resource(tic->pipe.texture);
if (tic->id < 0) {
tic->id = nvc0_screen_tic_alloc(nvc0->screen, tic);
PUSH_SPACE(push, 17);
BEGIN_NVC0(push, NVC0_M2MF(OFFSET_OUT_HIGH), 2);
PUSH_DATAh(push, txc->offset + (tic->id * 32));
PUSH_DATA (push, txc->offset + (tic->id * 32));
BEGIN_NVC0(push, NVC0_M2MF(LINE_LENGTH_IN), 2);
PUSH_DATA (push, 32);
PUSH_DATA (push, 1);
BEGIN_NVC0(push, NVC0_M2MF(EXEC), 1);
PUSH_DATA (push, 0x100111);
BEGIN_NIC0(push, NVC0_M2MF(DATA), 8);
PUSH_DATAp(push, &tic->tic[0], 8);
need_flush = TRUE;
} else
if (res->status & NOUVEAU_BUFFER_STATUS_GPU_WRITING) {
BEGIN_NVC0(push, NVC0_3D(TEX_CACHE_CTL), 1);
PUSH_DATA (push, (tic->id << 4) | 1);
}
nvc0->screen->tic.lock[tic->id / 32] |= 1 << (tic->id % 32);
res->status &= ~NOUVEAU_BUFFER_STATUS_GPU_WRITING;
res->status |= NOUVEAU_BUFFER_STATUS_GPU_READING;
if (!dirty)
continue;
commands[n++] = (tic->id << 9) | (i << 1) | 1;
BCTX_REFN(nvc0->bufctx_3d, TEX(s, i), res, RD);
}
for (; i < nvc0->state.num_textures[s]; ++i)
commands[n++] = (i << 1) | 0;
nvc0->state.num_textures[s] = nvc0->num_textures[s];
if (n) {
BEGIN_NIC0(push, NVC0_3D(BIND_TIC(s)), n);
PUSH_DATAp(push, commands, n);
}
nvc0->textures_dirty[s] = 0;
return need_flush;
}
static void
nvc0_push_upload_vertex_ids(struct push_context *ctx,
struct nvc0_context *nvc0,
const struct pipe_draw_info *info)
{
struct nouveau_pushbuf *push = ctx->push;
struct nouveau_bo *bo;
uint64_t va;
uint32_t *data;
uint32_t format;
unsigned index_size = nvc0->idxbuf.index_size;
unsigned i;
unsigned a = nvc0->vertex->num_elements;
if (!index_size || info->index_bias)
index_size = 4;
data = (uint32_t *)nouveau_scratch_get(&nvc0->base,
info->count * index_size, &va, &bo);
BCTX_REFN_bo(nvc0->bufctx_3d, VTX_TMP, NOUVEAU_BO_GART | NOUVEAU_BO_RD,
bo);
nouveau_pushbuf_validate(push);
if (info->indexed) {
if (!info->index_bias) {
memcpy(data, ctx->idxbuf, info->count * index_size);
} else {
switch (nvc0->idxbuf.index_size) {
case 1:
copy_indices_u8(data, ctx->idxbuf, info->index_bias, info->count);
break;
case 2:
copy_indices_u16(data, ctx->idxbuf, info->index_bias, info->count);
break;
default:
copy_indices_u32(data, ctx->idxbuf, info->index_bias, info->count);
break;
}
}
} else {
for (i = 0; i < info->count; ++i)
data[i] = i + (info->start + info->index_bias);
}
format = (1 << NVC0_3D_VERTEX_ATTRIB_FORMAT_BUFFER__SHIFT) |
NVC0_3D_VERTEX_ATTRIB_FORMAT_TYPE_UINT;
switch (index_size) {
case 1:
format |= NVC0_3D_VERTEX_ATTRIB_FORMAT_SIZE_8;
break;
case 2:
format |= NVC0_3D_VERTEX_ATTRIB_FORMAT_SIZE_16;
break;
default:
format |= NVC0_3D_VERTEX_ATTRIB_FORMAT_SIZE_32;
break;
}
PUSH_SPACE(push, 12);
if (unlikely(nvc0->state.instance_elts & 2)) {
nvc0->state.instance_elts &= ~2;
IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_PER_INSTANCE(1)), 0);
}
BEGIN_NVC0(push, NVC0_3D(VERTEX_ATTRIB_FORMAT(a)), 1);
PUSH_DATA (push, format);
BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(1)), 3);
PUSH_DATA (push, NVC0_3D_VERTEX_ARRAY_FETCH_ENABLE | index_size);
PUSH_DATAh(push, va);
PUSH_DATA (push, va);
BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_LIMIT_HIGH(1)), 2);
PUSH_DATAh(push, va + info->count * index_size - 1);
PUSH_DATA (push, va + info->count * index_size - 1);
#define NVC0_3D_VERTEX_ID_REPLACE_SOURCE_ATTR_X(a) \
(((0x80 + (a) * 0x10) / 4) << NVC0_3D_VERTEX_ID_REPLACE_SOURCE__SHIFT)
BEGIN_NVC0(push, NVC0_3D(VERTEX_ID_REPLACE), 1);
PUSH_DATA (push, NVC0_3D_VERTEX_ID_REPLACE_SOURCE_ATTR_X(a) | 1);
}
void
nvc0_push_vbo(struct nvc0_context *nvc0, const struct pipe_draw_info *info)
{
struct push_context ctx;
unsigned i, index_size;
unsigned inst_count = info->instance_count;
unsigned vert_count = info->count;
unsigned prim;
nvc0_push_context_init(nvc0, &ctx);
nvc0_vertex_configure_translate(nvc0, info->index_bias);
if (unlikely(ctx.edgeflag.enabled))
nvc0_push_map_edgeflag(&ctx, nvc0, info->index_bias);
ctx.prim_restart = info->primitive_restart;
ctx.restart_index = info->restart_index;
if (info->indexed) {
nvc0_push_map_idxbuf(&ctx, nvc0);
index_size = nvc0->idxbuf.index_size;
if (info->primitive_restart) {
BEGIN_NVC0(ctx.push, NVC0_3D(PRIM_RESTART_ENABLE), 2);
PUSH_DATA (ctx.push, 1);
PUSH_DATA (ctx.push, info->restart_index);
} else
if (nvc0->state.prim_restart) {
IMMED_NVC0(ctx.push, NVC0_3D(PRIM_RESTART_ENABLE), 0);
}
nvc0->state.prim_restart = info->primitive_restart;
} else {
if (unlikely(info->count_from_stream_output)) {
struct pipe_context *pipe = &nvc0->base.pipe;
struct nvc0_so_target *targ;
targ = nvc0_so_target(info->count_from_stream_output);
pipe->get_query_result(pipe, targ->pq, TRUE, (void *)&vert_count);
vert_count /= targ->stride;
}
ctx.idxbuf = NULL; /* shut up warnings */
index_size = 0;
}
ctx.instance_id = info->start_instance;
prim = nvc0_prim_gl(info->mode);
do {
PUSH_SPACE(ctx.push, 9);
ctx.dest = nvc0_push_setup_vertex_array(nvc0, vert_count);
if (unlikely(!ctx.dest))
break;
if (unlikely(ctx.need_vertex_id))
nvc0_push_upload_vertex_ids(&ctx, nvc0, info);
IMMED_NVC0(ctx.push, NVC0_3D(VERTEX_ARRAY_FLUSH), 0);
BEGIN_NVC0(ctx.push, NVC0_3D(VERTEX_BEGIN_GL), 1);
PUSH_DATA (ctx.push, prim);
switch (index_size) {
case 1:
disp_vertices_i08(&ctx, info->start, vert_count);
break;
case 2:
disp_vertices_i16(&ctx, info->start, vert_count);
break;
case 4:
disp_vertices_i32(&ctx, info->start, vert_count);
break;
default:
assert(index_size == 0);
disp_vertices_seq(&ctx, info->start, vert_count);
break;
}
PUSH_SPACE(ctx.push, 1);
IMMED_NVC0(ctx.push, NVC0_3D(VERTEX_END_GL), 0);
if (--inst_count) {
prim |= NVC0_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT;
++ctx.instance_id;
}
nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_VTX_TMP);
nouveau_scratch_done(&nvc0->base);
} while (inst_count);
/* reset state and unmap buffers (no-op) */
if (unlikely(!ctx.edgeflag.value)) {
PUSH_SPACE(ctx.push, 1);
IMMED_NVC0(ctx.push, NVC0_3D(EDGEFLAG), 1);
}
if (unlikely(ctx.need_vertex_id)) {
PUSH_SPACE(ctx.push, 4);
IMMED_NVC0(ctx.push, NVC0_3D(VERTEX_ID_REPLACE), 0);
BEGIN_NVC0(ctx.push, NVC0_3D(VERTEX_ATTRIB_FORMAT(1)), 1);
PUSH_DATA (ctx.push,
NVC0_3D_VERTEX_ATTRIB_FORMAT_CONST |
NVC0_3D_VERTEX_ATTRIB_FORMAT_TYPE_FLOAT |
NVC0_3D_VERTEX_ATTRIB_FORMAT_SIZE_32);
IMMED_NVC0(ctx.push, NVC0_3D(VERTEX_ARRAY_FETCH(1)), 0);
}
if (info->indexed)
nouveau_resource_unmap(nv04_resource(nvc0->idxbuf.buffer));
for (i = 0; i < nvc0->num_vtxbufs; ++i)
nouveau_resource_unmap(nv04_resource(nvc0->vtxbuf[i].buffer));
NOUVEAU_DRV_STAT(&nvc0->screen->base, draw_calls_fallback_count, 1);
}
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 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;
}
void
nvc0_fragprog_validate(struct nvc0_context *nvc0)
{
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
struct nvc0_program *fp = nvc0->fragprog;
struct pipe_rasterizer_state *rast = &nvc0->rast->pipe;
if (fp->fp.force_persample_interp != rast->force_persample_interp) {
/* Force the program to be reuploaded, which will trigger interp fixups
* to get applied
*/
if (fp->mem)
nouveau_heap_free(&fp->mem);
fp->fp.force_persample_interp = rast->force_persample_interp;
}
/* Shade model works well enough when both colors follow it. However if one
* (or both) is explicitly set, then we have to go the patching route.
*/
bool has_explicit_color = fp->fp.colors &&
(((fp->fp.colors & 1) && !fp->fp.color_interp[0]) ||
((fp->fp.colors & 2) && !fp->fp.color_interp[1]));
bool hwflatshade = false;
if (has_explicit_color && fp->fp.flatshade != rast->flatshade) {
/* Force re-upload */
if (fp->mem)
nouveau_heap_free(&fp->mem);
fp->fp.flatshade = rast->flatshade;
/* Always smooth-shade in this mode, the shader will decide on its own
* when to flat-shade.
*/
} else if (!has_explicit_color) {
hwflatshade = rast->flatshade;
/* No need to binary-patch the shader each time, make sure that it's set
* up for the default behaviour.
*/
fp->fp.flatshade = 0;
}
if (hwflatshade != nvc0->state.flatshade) {
nvc0->state.flatshade = hwflatshade;
BEGIN_NVC0(push, NVC0_3D(SHADE_MODEL), 1);
PUSH_DATA (push, hwflatshade ? NVC0_3D_SHADE_MODEL_FLAT :
NVC0_3D_SHADE_MODEL_SMOOTH);
}
if (fp->mem && !(nvc0->dirty_3d & NVC0_NEW_3D_FRAGPROG)) {
return;
}
if (!nvc0_program_validate(nvc0, fp))
return;
nvc0_program_update_context_state(nvc0, fp, 4);
if (fp->fp.early_z != nvc0->state.early_z_forced) {
nvc0->state.early_z_forced = fp->fp.early_z;
IMMED_NVC0(push, NVC0_3D(FORCE_EARLY_FRAGMENT_TESTS), fp->fp.early_z);
}
BEGIN_NVC0(push, NVC0_3D(SP_SELECT(5)), 2);
PUSH_DATA (push, 0x51);
PUSH_DATA (push, fp->code_base);
BEGIN_NVC0(push, NVC0_3D(SP_GPR_ALLOC(5)), 1);
PUSH_DATA (push, fp->num_gprs);
BEGIN_NVC0(push, SUBC_3D(0x0360), 2);
PUSH_DATA (push, 0x20164010);
PUSH_DATA (push, 0x20);
BEGIN_NVC0(push, NVC0_3D(ZCULL_TEST_MASK), 1);
PUSH_DATA (push, fp->flags[0]);
}
void
nvc0_push_vbo(struct nvc0_context *nvc0, const struct pipe_draw_info *info)
{
struct push_context ctx;
unsigned i, index_size;
unsigned inst_count = info->instance_count;
unsigned vert_count = info->count;
unsigned prim;
nvc0_push_context_init(nvc0, &ctx);
nvc0_vertex_configure_translate(nvc0, info->index_bias);
if (nvc0->state.index_bias) {
/* this is already taken care of by translate */
IMMED_NVC0(ctx.push, NVC0_3D(VB_ELEMENT_BASE), 0);
nvc0->state.index_bias = 0;
}
if (unlikely(ctx.edgeflag.enabled))
nvc0_push_map_edgeflag(&ctx, nvc0, info->index_bias);
ctx.prim_restart = info->primitive_restart;
ctx.restart_index = info->restart_index;
if (info->primitive_restart) {
/* NOTE: I hope we won't ever need that last index (~0).
* If we do, we have to disable primitive restart here always and
* use END,BEGIN to restart. (XXX: would that affect PrimitiveID ?)
* We could also deactive PRIM_RESTART_WITH_DRAW_ARRAYS temporarily,
* and add manual restart to disp_vertices_seq.
*/
BEGIN_NVC0(ctx.push, NVC0_3D(PRIM_RESTART_ENABLE), 2);
PUSH_DATA (ctx.push, 1);
PUSH_DATA (ctx.push, info->indexed ? 0xffffffff : info->restart_index);
} else
if (nvc0->state.prim_restart) {
IMMED_NVC0(ctx.push, NVC0_3D(PRIM_RESTART_ENABLE), 0);
}
nvc0->state.prim_restart = info->primitive_restart;
if (info->indexed) {
nvc0_push_map_idxbuf(&ctx, nvc0);
index_size = nvc0->idxbuf.index_size;
} else {
if (unlikely(info->count_from_stream_output)) {
struct pipe_context *pipe = &nvc0->base.pipe;
struct nvc0_so_target *targ;
targ = nvc0_so_target(info->count_from_stream_output);
pipe->get_query_result(pipe, targ->pq, true, (void *)&vert_count);
vert_count /= targ->stride;
}
ctx.idxbuf = NULL; /* shut up warnings */
index_size = 0;
}
ctx.instance_id = info->start_instance;
prim = nvc0_prim_gl(info->mode);
do {
PUSH_SPACE(ctx.push, 9);
ctx.dest = nvc0_push_setup_vertex_array(nvc0, vert_count);
if (unlikely(!ctx.dest))
break;
if (unlikely(ctx.need_vertex_id))
nvc0_push_upload_vertex_ids(&ctx, nvc0, info);
if (nvc0->screen->eng3d->oclass < GM107_3D_CLASS)
IMMED_NVC0(ctx.push, NVC0_3D(VERTEX_ARRAY_FLUSH), 0);
BEGIN_NVC0(ctx.push, NVC0_3D(VERTEX_BEGIN_GL), 1);
PUSH_DATA (ctx.push, prim);
switch (index_size) {
case 1:
disp_vertices_i08(&ctx, info->start, vert_count);
break;
case 2:
disp_vertices_i16(&ctx, info->start, vert_count);
break;
case 4:
disp_vertices_i32(&ctx, info->start, vert_count);
break;
default:
assert(index_size == 0);
disp_vertices_seq(&ctx, info->start, vert_count);
break;
}
PUSH_SPACE(ctx.push, 1);
IMMED_NVC0(ctx.push, NVC0_3D(VERTEX_END_GL), 0);
if (--inst_count) {
prim |= NVC0_3D_VERTEX_BEGIN_GL_INSTANCE_NEXT;
++ctx.instance_id;
}
nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_VTX_TMP);
nouveau_scratch_done(&nvc0->base);
} while (inst_count);
/* reset state and unmap buffers (no-op) */
if (unlikely(!ctx.edgeflag.value)) {
PUSH_SPACE(ctx.push, 1);
IMMED_NVC0(ctx.push, NVC0_3D(EDGEFLAG), 1);
}
if (unlikely(ctx.need_vertex_id)) {
PUSH_SPACE(ctx.push, 4);
IMMED_NVC0(ctx.push, NVC0_3D(VERTEX_ID_REPLACE), 0);
BEGIN_NVC0(ctx.push, NVC0_3D(VERTEX_ATTRIB_FORMAT(1)), 1);
PUSH_DATA (ctx.push,
NVC0_3D_VERTEX_ATTRIB_FORMAT_CONST |
NVC0_3D_VERTEX_ATTRIB_FORMAT_TYPE_FLOAT |
NVC0_3D_VERTEX_ATTRIB_FORMAT_SIZE_32);
IMMED_NVC0(ctx.push, NVC0_3D(VERTEX_ARRAY_FETCH(1)), 0);
}
if (info->indexed)
nouveau_resource_unmap(nv04_resource(nvc0->idxbuf.buffer));
for (i = 0; i < nvc0->num_vtxbufs; ++i)
nouveau_resource_unmap(nv04_resource(nvc0->vtxbuf[i].buffer));
NOUVEAU_DRV_STAT(&nvc0->screen->base, draw_calls_fallback_count, 1);
}