static void
nvc0_set_transform_feedback_targets(struct pipe_context *pipe,
unsigned num_targets,
struct pipe_stream_output_target **targets,
unsigned append_mask)
{
struct nvc0_context *nvc0 = nvc0_context(pipe);
unsigned i;
boolean serialize = TRUE;
assert(num_targets <= 4);
for (i = 0; i < num_targets; ++i) {
if (nvc0->tfbbuf[i] == targets[i] && (append_mask & (1 << i)))
continue;
nvc0->tfbbuf_dirty |= 1 << i;
if (nvc0->tfbbuf[i] && nvc0->tfbbuf[i] != targets[i])
nvc0_so_target_save_offset(pipe, nvc0->tfbbuf[i], i, &serialize);
if (targets[i] && !(append_mask & (1 << i)))
nvc0_so_target(targets[i])->clean = TRUE;
pipe_so_target_reference(&nvc0->tfbbuf[i], targets[i]);
}
for (; i < nvc0->num_tfbbufs; ++i) {
nvc0->tfbbuf_dirty |= 1 << i;
nvc0_so_target_save_offset(pipe, nvc0->tfbbuf[i], i, &serialize);
pipe_so_target_reference(&nvc0->tfbbuf[i], NULL);
}
nvc0->num_tfbbufs = num_targets;
if (nvc0->tfbbuf_dirty)
nvc0->dirty |= NVC0_NEW_TFB_TARGETS;
}
static void
nvc0_so_target_destroy(struct pipe_context *pipe,
struct pipe_stream_output_target *ptarg)
{
struct nvc0_so_target *targ = nvc0_so_target(ptarg);
pipe->destroy_query(pipe, targ->pq);
pipe_resource_reference(&targ->pipe.buffer, NULL);
FREE(targ);
}
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);
}
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_set_transform_feedback_targets(struct pipe_context *pipe,
unsigned num_targets,
struct pipe_stream_output_target **targets,
const unsigned *offsets)
{
struct nvc0_context *nvc0 = nvc0_context(pipe);
unsigned i;
bool serialize = true;
assert(num_targets <= 4);
for (i = 0; i < num_targets; ++i) {
const bool changed = nvc0->tfbbuf[i] != targets[i];
const bool append = (offsets[i] == ((unsigned)-1));
if (!changed && append)
continue;
nvc0->tfbbuf_dirty |= 1 << i;
if (nvc0->tfbbuf[i] && changed)
nvc0_so_target_save_offset(pipe, nvc0->tfbbuf[i], i, &serialize);
if (targets[i] && !append)
nvc0_so_target(targets[i])->clean = true;
pipe_so_target_reference(&nvc0->tfbbuf[i], targets[i]);
}
for (; i < nvc0->num_tfbbufs; ++i) {
if (nvc0->tfbbuf[i]) {
nvc0->tfbbuf_dirty |= 1 << i;
nvc0_so_target_save_offset(pipe, nvc0->tfbbuf[i], i, &serialize);
pipe_so_target_reference(&nvc0->tfbbuf[i], NULL);
}
}
nvc0->num_tfbbufs = num_targets;
if (nvc0->tfbbuf_dirty) {
nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_3D_TFB);
nvc0->dirty_3d |= NVC0_NEW_3D_TFB_TARGETS;
}
}
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);
}
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);
}
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);
}
void
nvc0_tfb_validate(struct nvc0_context *nvc0)
{
struct nouveau_channel *chan = nvc0->screen->base.channel;
struct nvc0_transform_feedback_state *tfb;
unsigned b, n, i;
if (nvc0->gmtyprog) tfb = nvc0->gmtyprog->tfb;
else
if (nvc0->tevlprog) tfb = nvc0->tevlprog->tfb;
else
tfb = nvc0->vertprog->tfb;
IMMED_RING(chan, RING_3D(TFB_ENABLE), (tfb && nvc0->num_tfbbufs) ? 1 : 0);
if (tfb && tfb != nvc0->state.tfb) {
uint8_t var[128];
for (n = 0, b = 0; b < 4; n += tfb->varying_count[b++]) {
if (tfb->varying_count[b]) {
BEGIN_RING(chan, RING_3D(TFB_STREAM(b)), 3);
OUT_RING (chan, 0);
OUT_RING (chan, tfb->varying_count[b]);
OUT_RING (chan, tfb->stride[b]);
for (i = 0; i < tfb->varying_count[b]; ++i)
var[i] = tfb->varying_index[n + i];
for (; i & 3; ++i)
var[i] = 0; /* zero rest of method word bits */
BEGIN_RING(chan, RING_3D(TFB_VARYING_LOCS(b, 0)), i / 4);
OUT_RINGp (chan, var, i / 4);
if (nvc0->tfbbuf[b])
nvc0_so_target(nvc0->tfbbuf[b])->stride = tfb->stride[b];
} else {
IMMED_RING(chan, RING_3D(TFB_VARYING_COUNT(b)), 0);
}
}
}
nvc0->state.tfb = tfb;
if (!(nvc0->dirty & NVC0_NEW_TFB_TARGETS))
return;
nvc0_bufctx_reset(nvc0, NVC0_BUFCTX_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(chan, targ->pq);
BEGIN_RING(chan, RING_3D(TFB_BUFFER_ENABLE(b)), 5);
OUT_RING (chan, 1);
OUT_RESRCh(chan, buf, targ->pipe.buffer_offset, NOUVEAU_BO_WR);
OUT_RESRCl(chan, buf, targ->pipe.buffer_offset, NOUVEAU_BO_WR);
OUT_RING (chan, targ->pipe.buffer_size);
if (!targ->clean) {
nvc0_query_pushbuf_submit(chan, targ->pq, 0x4);
} else {
OUT_RING(chan, 0); /* TFB_BUFFER_OFFSET */
targ->clean = FALSE;
}
nvc0_bufctx_add_resident(nvc0, NVC0_BUFCTX_TFB, buf, NOUVEAU_BO_WR);
}
for (; b < 4; ++b)
IMMED_RING(chan, RING_3D(TFB_BUFFER_ENABLE(b)), 0);
}