struct nvc0_vertex_stateobj *so;
struct translate_key transkey;
unsigned i;
+ unsigned src_offset_max = 0;
so = MALLOC(sizeof(*so) +
num_elements * sizeof(struct nvc0_vertex_element));
so->num_elements = num_elements;
so->instance_elts = 0;
so->instance_bufs = 0;
+ so->shared_slots = FALSE;
so->need_conversion = FALSE;
+ memset(so->vb_access_size, 0, sizeof(so->vb_access_size));
+
+ for (i = 0; i < PIPE_MAX_ATTRIBS; ++i)
+ so->min_instance_div[i] = 0xffffffff;
+
transkey.nr_elements = 0;
transkey.output_stride = 0;
for (i = 0; i < num_elements; ++i) {
const struct pipe_vertex_element *ve = &elements[i];
const unsigned vbi = ve->vertex_buffer_index;
+ unsigned size;
enum pipe_format fmt = ve->src_format;
so->element[i].pipe = elements[i];
so->element[i].state = nvc0_format_table[fmt].vtx;
so->need_conversion = TRUE;
}
- so->element[i].state |= i;
+ size = util_format_get_blocksize(fmt);
+
+ src_offset_max = MAX2(src_offset_max, ve->src_offset);
+
+ if (so->vb_access_size[vbi] < (ve->src_offset + size))
+ so->vb_access_size[vbi] = ve->src_offset + size;
+
+ if (unlikely(ve->instance_divisor)) {
+ so->instance_elts |= 1 << i;
+ so->instance_bufs |= 1 << vbi;
+ if (ve->instance_divisor < so->min_instance_div[vbi])
+ so->min_instance_div[vbi] = ve->instance_divisor;
+ }
if (1) {
+ unsigned ca;
unsigned j = transkey.nr_elements++;
+ ca = util_format_description(fmt)->channel[0].size / 8;
+ if (ca != 1 && ca != 2)
+ ca = 4;
+
transkey.element[j].type = TRANSLATE_ELEMENT_NORMAL;
transkey.element[j].input_format = ve->src_format;
transkey.element[j].input_buffer = vbi;
transkey.element[j].input_offset = ve->src_offset;
transkey.element[j].instance_divisor = ve->instance_divisor;
+ transkey.output_stride = align(transkey.output_stride, ca);
transkey.element[j].output_format = fmt;
transkey.element[j].output_offset = transkey.output_stride;
- transkey.output_stride += (util_format_get_stride(fmt, 1) + 3) & ~3;
+ transkey.output_stride += size;
- if (unlikely(ve->instance_divisor)) {
- so->instance_elts |= 1 << i;
- so->instance_bufs |= 1 << vbi;
- }
+ so->element[i].state_alt = so->element[i].state;
+ so->element[i].state_alt |= transkey.element[j].output_offset << 7;
}
+
+ so->element[i].state |= i << NVC0_3D_VERTEX_ATTRIB_FORMAT_BUFFER__SHIFT;
}
+ transkey.output_stride = align(transkey.output_stride, 4);
+ so->size = transkey.output_stride;
so->translate = translate_create(&transkey);
- so->vtx_size = transkey.output_stride / 4;
- so->vtx_per_packet_max = NV04_PFIFO_MAX_PACKET_LEN / MAX2(so->vtx_size, 1);
+ if (so->instance_elts || src_offset_max >= (1 << 14))
+ return so;
+ so->shared_slots = TRUE;
+
+ for (i = 0; i < num_elements; ++i) {
+ const unsigned b = elements[i].vertex_buffer_index;
+ const unsigned s = elements[i].src_offset;
+ so->element[i].state &= ~NVC0_3D_VERTEX_ATTRIB_FORMAT_BUFFER__MASK;
+ so->element[i].state |= b << NVC0_3D_VERTEX_ATTRIB_FORMAT_BUFFER__SHIFT;
+ so->element[i].state |= s << NVC0_3D_VERTEX_ATTRIB_FORMAT_OFFSET__SHIFT;
+ }
return so;
}
((c) << NVC0_3D_VTX_ATTR_DEFINE_COMP__SHIFT))
static void
-nvc0_emit_vtxattr(struct nvc0_context *nvc0, struct pipe_vertex_buffer *vb,
- struct pipe_vertex_element *ve, unsigned attr)
+nvc0_set_constant_vertex_attrib(struct nvc0_context *nvc0, const unsigned a)
{
- const void *data;
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
- struct nv04_resource *res = nv04_resource(vb->buffer);
- float v[4];
- int i;
- const unsigned nc = util_format_get_nr_components(ve->src_format);
+ struct pipe_vertex_element *ve = &nvc0->vertex->element[a].pipe;
+ struct pipe_vertex_buffer *vb = &nvc0->vtxbuf[ve->vertex_buffer_index];
+ uint32_t mode;
+ const struct util_format_description *desc;
+ void *dst;
+ const void *src = (const uint8_t *)vb->user_buffer + ve->src_offset;
+ assert(!vb->buffer);
- data = nouveau_resource_map_offset(&nvc0->base, res, vb->buffer_offset +
- ve->src_offset, NOUVEAU_BO_RD);
-
- util_format_read_4f(ve->src_format, v, 0, data, 0, 0, 0, 1, 1);
+ desc = util_format_description(ve->src_format);
PUSH_SPACE(push, 6);
- BEGIN_NVC0(push, NVC0_3D(VTX_ATTR_DEFINE), nc + 1);
- PUSH_DATA (push, VTX_ATTR(attr, nc, FLOAT, 32));
- for (i = 0; i < nc; ++i)
- PUSH_DATAf(push, v[i]);
+ BEGIN_NVC0(push, NVC0_3D(VTX_ATTR_DEFINE), 5);
+ dst = &push->cur[1];
+ if (desc->channel[0].pure_integer) {
+ if (desc->channel[0].type == UTIL_FORMAT_TYPE_SIGNED) {
+ mode = VTX_ATTR(a, 4, SINT, 32);
+ desc->unpack_rgba_sint(dst, 0, src, 0, 1, 1);
+ } else {
+ mode = VTX_ATTR(a, 4, UINT, 32);
+ desc->unpack_rgba_uint(dst, 0, src, 0, 1, 1);
+ }
+ } else {
+ mode = VTX_ATTR(a, 4, FLOAT, 32);
+ desc->unpack_rgba_float(dst, 0, src, 0, 1, 1);
+ }
+ push->cur[0] = mode;
+ push->cur += 5;
}
static INLINE void
-nvc0_vbuf_range(struct nvc0_context *nvc0, int vbi,
- uint32_t *base, uint32_t *size)
+nvc0_user_vbuf_range(struct nvc0_context *nvc0, int vbi,
+ uint32_t *base, uint32_t *size)
{
if (unlikely(nvc0->vertex->instance_bufs & (1 << vbi))) {
- /* TODO: use min and max instance divisor to get a proper range */
- *base = 0;
- *size = nvc0->vtxbuf[vbi].buffer->width0;
+ const uint32_t div = nvc0->vertex->min_instance_div[vbi];
+ *base = nvc0->instance_off * nvc0->vtxbuf[vbi].stride;
+ *size = (nvc0->instance_max / div) * nvc0->vtxbuf[vbi].stride +
+ nvc0->vertex->vb_access_size[vbi];
} else {
- assert(nvc0->vbo_max_index != ~0);
- *base = nvc0->vbo_min_index * nvc0->vtxbuf[vbi].stride;
- *size = (nvc0->vbo_max_index -
- nvc0->vbo_min_index + 1) * nvc0->vtxbuf[vbi].stride;
+ /* NOTE: if there are user buffers, we *must* have index bounds */
+ assert(nvc0->vb_elt_limit != ~0);
+ *base = nvc0->vb_elt_first * nvc0->vtxbuf[vbi].stride;
+ *size = nvc0->vb_elt_limit * nvc0->vtxbuf[vbi].stride +
+ nvc0->vertex->vb_access_size[vbi];
}
}
-static void
-nvc0_prevalidate_vbufs(struct nvc0_context *nvc0)
+static INLINE void
+nvc0_release_user_vbufs(struct nvc0_context *nvc0)
{
- const uint32_t bo_flags = NOUVEAU_BO_RD | NOUVEAU_BO_GART;
- struct nouveau_bo *bo;
- struct pipe_vertex_buffer *vb;
- struct nv04_resource *buf;
- int i;
- uint32_t base, size;
-
- nvc0->vbo_fifo = nvc0->vbo_user = 0;
-
- nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_VTX);
-
- for (i = 0; i < nvc0->num_vtxbufs; ++i) {
- vb = &nvc0->vtxbuf[i];
- if (!vb->stride)
- continue;
- buf = nv04_resource(vb->buffer);
-
- if (!nouveau_resource_mapped_by_gpu(vb->buffer)) {
- if (nvc0->vbo_push_hint) {
- nvc0->vbo_fifo = ~0;
- return;
- }
- nvc0->base.vbo_dirty = TRUE;
-
- if (buf->status & NOUVEAU_BUFFER_STATUS_USER_MEMORY) {
- assert(vb->stride > vb->buffer_offset);
- nvc0->vbo_user |= 1 << i;
- nvc0_vbuf_range(nvc0, i, &base, &size);
- bo = nouveau_scratch_data(&nvc0->base, buf, base, size);
- if (bo)
- BCTX_REFN_bo(nvc0->bufctx_3d, VTX_TMP, bo_flags, bo);
- continue;
- } else {
- nouveau_buffer_migrate(&nvc0->base, buf, NOUVEAU_BO_GART);
- }
- }
- BCTX_REFN(nvc0->bufctx_3d, VTX, buf, RD);
+ if (nvc0->vbo_user) {
+ nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_VTX_TMP);
+ nouveau_scratch_done(&nvc0->base);
}
}
static void
nvc0_update_user_vbufs(struct nvc0_context *nvc0)
{
- const uint32_t bo_flags = NOUVEAU_BO_RD | NOUVEAU_BO_GART;
- struct nouveau_bo *bo;
+ uint64_t address[PIPE_MAX_ATTRIBS];
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
- uint32_t base, offset, size;
int i;
uint32_t written = 0;
PUSH_SPACE(push, nvc0->vertex->num_elements * 8);
-
for (i = 0; i < nvc0->vertex->num_elements; ++i) {
struct pipe_vertex_element *ve = &nvc0->vertex->element[i].pipe;
- const int b = ve->vertex_buffer_index;
+ const unsigned b = ve->vertex_buffer_index;
struct pipe_vertex_buffer *vb = &nvc0->vtxbuf[b];
- struct nv04_resource *buf = nv04_resource(vb->buffer);
+ uint32_t base, size;
if (!(nvc0->vbo_user & (1 << b)))
continue;
-
if (!vb->stride) {
- nvc0_emit_vtxattr(nvc0, vb, ve, i);
+ nvc0_set_constant_vertex_attrib(nvc0, i);
continue;
}
- nvc0_vbuf_range(nvc0, b, &base, &size);
+ nvc0_user_vbuf_range(nvc0, b, &base, &size);
if (!(written & (1 << b))) {
+ struct nouveau_bo *bo;
+ const uint32_t bo_flags = NOUVEAU_BO_RD | NOUVEAU_BO_GART;
written |= 1 << b;
- bo = nouveau_scratch_data(&nvc0->base, buf, base, size);
+ address[b] = nouveau_scratch_data(&nvc0->base, vb->user_buffer,
+ base, size, &bo);
if (bo)
BCTX_REFN_bo(nvc0->bufctx_3d, VTX_TMP, bo_flags, bo);
}
- offset = vb->buffer_offset + ve->src_offset;
- BEGIN_1IC0(push, NVC0_3D(VERTEX_ARRAY_SELECT), 5);
+ BEGIN_1IC0(push, NVC0_3D(MACRO_VERTEX_ARRAY_SELECT), 5);
PUSH_DATA (push, i);
- PUSH_DATAh(push, buf->address + base + size - 1);
- PUSH_DATA (push, buf->address + base + size - 1);
- PUSH_DATAh(push, buf->address + offset);
- PUSH_DATA (push, buf->address + offset);
+ PUSH_DATAh(push, address[b] + base + size - 1);
+ PUSH_DATA (push, address[b] + base + size - 1);
+ PUSH_DATAh(push, address[b] + ve->src_offset);
+ PUSH_DATA (push, address[b] + ve->src_offset);
}
nvc0->base.vbo_dirty = TRUE;
}
-static INLINE void
-nvc0_release_user_vbufs(struct nvc0_context *nvc0)
+static void
+nvc0_update_user_vbufs_shared(struct nvc0_context *nvc0)
{
- if (nvc0->vbo_user) {
- nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_VTX_TMP);
- nouveau_scratch_done(&nvc0->base);
+ struct nouveau_pushbuf *push = nvc0->base.pushbuf;
+ uint32_t mask = nvc0->vbo_user & ~nvc0->constant_vbos;
+
+ PUSH_SPACE(push, nvc0->num_vtxbufs * 8);
+ while (mask) {
+ struct nouveau_bo *bo;
+ const uint32_t bo_flags = NOUVEAU_BO_RD | NOUVEAU_BO_GART;
+ uint64_t address;
+ uint32_t base, size;
+ const int b = ffs(mask) - 1;
+ mask &= ~(1 << b);
+
+ nvc0_user_vbuf_range(nvc0, b, &base, &size);
+
+ address = nouveau_scratch_data(&nvc0->base, nvc0->vtxbuf[b].user_buffer,
+ base, size, &bo);
+ if (bo)
+ BCTX_REFN_bo(nvc0->bufctx_3d, VTX_TMP, bo_flags, bo);
+
+ BEGIN_1IC0(push, NVC0_3D(MACRO_VERTEX_ARRAY_SELECT), 5);
+ PUSH_DATA (push, b);
+ PUSH_DATAh(push, address + base + size - 1);
+ PUSH_DATA (push, address + base + size - 1);
+ PUSH_DATAh(push, address);
+ PUSH_DATA (push, address);
+ }
+
+ mask = nvc0->state.constant_elts;
+ while (mask) {
+ int i = ffs(mask) - 1;
+ mask &= ~(1 << i);
+ nvc0_set_constant_vertex_attrib(nvc0, i);
}
}
-void
-nvc0_vertex_arrays_validate(struct nvc0_context *nvc0)
+static void
+nvc0_validate_vertex_buffers(struct nvc0_context *nvc0)
{
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
- struct nvc0_vertex_stateobj *vertex = nvc0->vertex;
- struct pipe_vertex_buffer *vb;
- struct nvc0_vertex_element *ve;
+ const struct nvc0_vertex_stateobj *vertex = nvc0->vertex;
+ uint32_t refd = 0;
unsigned i;
- if (unlikely(vertex->need_conversion) ||
- unlikely(nvc0->vertprog->vp.edgeflag < PIPE_MAX_ATTRIBS)) {
- nvc0->vbo_fifo = ~0;
- nvc0->vbo_user = 0;
- } else {
- nvc0_prevalidate_vbufs(nvc0);
- }
-
- PUSH_SPACE(push, vertex->num_elements + 1);
- BEGIN_NVC0(push, NVC0_3D(VERTEX_ATTRIB_FORMAT(0)), vertex->num_elements);
+ PUSH_SPACE(push, vertex->num_elements * 8);
for (i = 0; i < vertex->num_elements; ++i) {
+ const struct nvc0_vertex_element *ve;
+ const struct pipe_vertex_buffer *vb;
+ struct nv04_resource *res;
+ unsigned b;
+ unsigned limit, offset;
+
+ if (nvc0->state.constant_elts & (1 << i))
+ continue;
ve = &vertex->element[i];
- vb = &nvc0->vtxbuf[ve->pipe.vertex_buffer_index];
+ b = ve->pipe.vertex_buffer_index;
+ vb = &nvc0->vtxbuf[b];
+
+ if (!vb->buffer) {
+ if (vb->stride) {
+ if (ve->pipe.instance_divisor) {
+ BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_DIVISOR(i)), 1);
+ PUSH_DATA (push, ve->pipe.instance_divisor);
+ }
+ BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(i)), 1);
+ PUSH_DATA (push, (1 << 12) | vb->stride);
+ }
+ /* address/value set in nvc0_update_user_vbufs */
+ continue;
+ }
+ res = nv04_resource(vb->buffer);
+ offset = ve->pipe.src_offset + vb->buffer_offset;
+ limit = vb->buffer->width0 - 1;
- if (likely(vb->stride) || nvc0->vbo_fifo) {
- PUSH_DATA(push, ve->state);
+ if (unlikely(ve->pipe.instance_divisor)) {
+ BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(i)), 4);
+ PUSH_DATA (push, (1 << 12) | vb->stride);
+ PUSH_DATAh(push, res->address + offset);
+ PUSH_DATA (push, res->address + offset);
+ PUSH_DATA (push, ve->pipe.instance_divisor);
} else {
- PUSH_DATA(push, ve->state | NVC0_3D_VERTEX_ATTRIB_FORMAT_CONST);
- nvc0->vbo_fifo &= ~(1 << i);
+ BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(i)), 3);
+ PUSH_DATA (push, (1 << 12) | vb->stride);
+ PUSH_DATAh(push, res->address + offset);
+ PUSH_DATA (push, res->address + offset);
+ }
+ BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_LIMIT_HIGH(i)), 2);
+ PUSH_DATAh(push, res->address + limit);
+ PUSH_DATA (push, res->address + limit);
+
+ if (!(refd & (1 << b))) {
+ refd |= 1 << b;
+ BCTX_REFN(nvc0->bufctx_3d, VTX, res, RD);
}
}
+ if (nvc0->vbo_user)
+ nvc0_update_user_vbufs(nvc0);
+}
- PUSH_SPACE(push, vertex->num_elements * 16);
- for (i = 0; i < vertex->num_elements; ++i) {
- struct nv04_resource *res;
- unsigned size, offset;
-
- ve = &vertex->element[i];
- vb = &nvc0->vtxbuf[ve->pipe.vertex_buffer_index];
+static void
+nvc0_validate_vertex_buffers_shared(struct nvc0_context *nvc0)
+{
+ struct nouveau_pushbuf *push = nvc0->base.pushbuf;
+ unsigned b;
+ const uint32_t mask = nvc0->vbo_user;
- if (unlikely(ve->pipe.instance_divisor)) {
- if (!(nvc0->state.instance_elts & (1 << i))) {
- IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_PER_INSTANCE(i)), 1);
+ PUSH_SPACE(push, nvc0->num_vtxbufs * 8);
+ for (b = 0; b < nvc0->num_vtxbufs; ++b) {
+ struct pipe_vertex_buffer *vb = &nvc0->vtxbuf[b];
+ struct nv04_resource *buf;
+ uint32_t offset, limit;
+
+ if (mask & (1 << b)) {
+ if (vb->stride) {
+ BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(b)), 1);
+ PUSH_DATA (push, NVC0_3D_VERTEX_ARRAY_FETCH_ENABLE | vb->stride);
}
- BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_DIVISOR(i)), 1);
- PUSH_DATA (push, ve->pipe.instance_divisor);
- } else
- if (unlikely(nvc0->state.instance_elts & (1 << i))) {
- IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_PER_INSTANCE(i)), 0);
+ /* address/value set in nvc0_update_user_vbufs_shared */
+ continue;
}
+ buf = nv04_resource(vb->buffer);
+ offset = vb->buffer_offset;
+ limit = buf->base.width0 - 1;
- res = nv04_resource(vb->buffer);
+ BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(b)), 3);
+ PUSH_DATA (push, NVC0_3D_VERTEX_ARRAY_FETCH_ENABLE | vb->stride);
+ PUSH_DATAh(push, buf->address + offset);
+ PUSH_DATA (push, buf->address + offset);
+ BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_LIMIT_HIGH(b)), 2);
+ PUSH_DATAh(push, buf->address + limit);
+ PUSH_DATA (push, buf->address + limit);
- if (nvc0->vbo_fifo || unlikely(vb->stride == 0)) {
- if (!nvc0->vbo_fifo)
- nvc0_emit_vtxattr(nvc0, vb, &ve->pipe, i);
- BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(i)), 1);
- PUSH_DATA (push, 0);
- continue;
- }
+ BCTX_REFN(nvc0->bufctx_3d, VTX, buf, RD);
+ }
+ if (nvc0->vbo_user)
+ nvc0_update_user_vbufs_shared(nvc0);
+}
- size = vb->buffer->width0;
- offset = ve->pipe.src_offset + vb->buffer_offset;
+void
+nvc0_vertex_arrays_validate(struct nvc0_context *nvc0)
+{
+ struct nouveau_pushbuf *push = nvc0->base.pushbuf;
+ struct nvc0_vertex_stateobj *vertex = nvc0->vertex;
+ struct nvc0_vertex_element *ve;
+ uint32_t const_vbos;
+ unsigned i;
+ uint8_t vbo_mode;
+ boolean update_vertex;
- BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(i)), 1);
- PUSH_DATA (push, (1 << 12) | vb->stride);
- BEGIN_1IC0(push, NVC0_3D(VERTEX_ARRAY_SELECT), 5);
- PUSH_DATA (push, i);
- PUSH_DATAh(push, res->address + size - 1);
- PUSH_DATA (push, res->address + size - 1);
- PUSH_DATAh(push, res->address + offset);
- PUSH_DATA (push, res->address + offset);
+ nouveau_bufctx_reset(nvc0->bufctx_3d, NVC0_BIND_VTX);
+
+ if (unlikely(vertex->need_conversion) ||
+ unlikely(nvc0->vertprog->vp.edgeflag < PIPE_MAX_ATTRIBS)) {
+ vbo_mode = 3;
+ } else {
+ vbo_mode = (nvc0->vbo_user && nvc0->vbo_push_hint) ? 1 : 0;
}
- for (; i < nvc0->state.num_vtxelts; ++i) {
- PUSH_SPACE(push, 5);
- BEGIN_NVC0(push, NVC0_3D(VERTEX_ATTRIB_FORMAT(i)), 1);
- PUSH_DATA (push, NVC0_3D_VERTEX_ATTRIB_INACTIVE);
- if (unlikely(nvc0->state.instance_elts & (1 << i)))
- IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_PER_INSTANCE(i)), 0);
- BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(i)), 1);
- PUSH_DATA (push, 0);
+ const_vbos = vbo_mode ? 0 : nvc0->constant_vbos;
+
+ update_vertex = (nvc0->dirty & NVC0_NEW_VERTEX) ||
+ (const_vbos != nvc0->state.constant_vbos) ||
+ (vbo_mode != nvc0->state.vbo_mode);
+
+ if (update_vertex) {
+ const unsigned n = MAX2(vertex->num_elements, nvc0->state.num_vtxelts);
+
+ nvc0->state.constant_vbos = const_vbos;
+ nvc0->state.constant_elts = 0;
+ nvc0->state.num_vtxelts = vertex->num_elements;
+ nvc0->state.vbo_mode = vbo_mode;
+
+ if (unlikely(vbo_mode)) {
+ if (unlikely(nvc0->state.instance_elts & 3)) {
+ /* translate mode uses only 2 vertex buffers */
+ nvc0->state.instance_elts &= ~3;
+ PUSH_SPACE(push, 3);
+ BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_PER_INSTANCE(0)), 2);
+ PUSH_DATA (push, 0);
+ PUSH_DATA (push, 0);
+ }
+
+ PUSH_SPACE(push, n * 2 + 4);
+
+ BEGIN_NVC0(push, NVC0_3D(VERTEX_ATTRIB_FORMAT(0)), n);
+ for (i = 0; i < vertex->num_elements; ++i)
+ PUSH_DATA(push, vertex->element[i].state_alt);
+ for (; i < n; ++i)
+ PUSH_DATA(push, NVC0_3D_VERTEX_ATTRIB_INACTIVE);
+
+ BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(0)), 1);
+ PUSH_DATA (push, (1 << 12) | vertex->size);
+ for (i = 1; i < n; ++i)
+ IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(i)), 0);
+ } else {
+ uint32_t *restrict data;
+
+ if (unlikely(vertex->instance_elts != nvc0->state.instance_elts)) {
+ nvc0->state.instance_elts = vertex->instance_elts;
+ assert(n); /* if (n == 0), both masks should be 0 */
+ PUSH_SPACE(push, 3);
+ BEGIN_NVC0(push, NVC0_3D(MACRO_VERTEX_ARRAY_PER_INSTANCE), 2);
+ PUSH_DATA (push, n);
+ PUSH_DATA (push, vertex->instance_elts);
+ }
+
+ PUSH_SPACE(push, n * 2 + 1);
+ BEGIN_NVC0(push, NVC0_3D(VERTEX_ATTRIB_FORMAT(0)), n);
+ data = push->cur;
+ push->cur += n;
+ for (i = 0; i < vertex->num_elements; ++i) {
+ ve = &vertex->element[i];
+ data[i] = ve->state;
+ if (unlikely(const_vbos & (1 << ve->pipe.vertex_buffer_index))) {
+ nvc0->state.constant_elts |= 1 << i;
+ data[i] |= NVC0_3D_VERTEX_ATTRIB_FORMAT_CONST;
+ IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(i)), 0);
+ }
+ }
+ for (; i < n; ++i) {
+ data[i] = NVC0_3D_VERTEX_ATTRIB_INACTIVE;
+ IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_FETCH(i)), 0);
+ }
+ }
}
+ if (nvc0->state.vbo_mode) /* using translate, don't set up arrays here */
+ return;
- nvc0->state.num_vtxelts = vertex->num_elements;
- nvc0->state.instance_elts = vertex->instance_elts;
+ if (vertex->shared_slots)
+ nvc0_validate_vertex_buffers_shared(nvc0);
+ else
+ nvc0_validate_vertex_buffers(nvc0);
}
void
struct nv04_resource *buf = nv04_resource(nvc0->idxbuf.buffer);
assert(buf);
- if (!nouveau_resource_mapped_by_gpu(&buf->base))
- return;
+ assert(nouveau_resource_mapped_by_gpu(&buf->base));
PUSH_SPACE(push, 6);
BEGIN_NVC0(push, NVC0_3D(INDEX_ARRAY_START_HIGH), 5);
NVC0_PRIM_GL_CASE(PATCHES); */
default:
return NVC0_3D_VERTEX_BEGIN_GL_PRIMITIVE_POINTS;
- break;
}
}
}
static void
-nvc0_draw_elements_inline_u08(struct nouveau_pushbuf *push, uint8_t *map,
+nvc0_draw_elements_inline_u08(struct nouveau_pushbuf *push, const uint8_t *map,
unsigned start, unsigned count)
{
map += start;
}
static void
-nvc0_draw_elements_inline_u16(struct nouveau_pushbuf *push, uint16_t *map,
+nvc0_draw_elements_inline_u16(struct nouveau_pushbuf *push, const uint16_t *map,
unsigned start, unsigned count)
{
map += start;
}
static void
-nvc0_draw_elements_inline_u32(struct nouveau_pushbuf *push, uint32_t *map,
+nvc0_draw_elements_inline_u32(struct nouveau_pushbuf *push, const uint32_t *map,
unsigned start, unsigned count)
{
map += start;
}
static void
-nvc0_draw_elements_inline_u32_short(struct nouveau_pushbuf *push, uint32_t *map,
+nvc0_draw_elements_inline_u32_short(struct nouveau_pushbuf *push,
+ const uint32_t *map,
unsigned start, unsigned count)
{
map += start;
unsigned instance_count, int32_t index_bias)
{
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
- void *data;
unsigned prim;
const unsigned index_size = nvc0->idxbuf.index_size;
nvc0->state.index_bias = index_bias;
}
- if (nouveau_resource_mapped_by_gpu(nvc0->idxbuf.buffer)) {
+ if (nvc0->idxbuf.buffer) {
PUSH_SPACE(push, 1);
IMMED_NVC0(push, NVC0_3D(VERTEX_BEGIN_GL), prim);
do {
} while (instance_count);
IMMED_NVC0(push, NVC0_3D(VERTEX_END_GL), 0);
} else {
- data = nouveau_resource_map_offset(&nvc0->base,
- nv04_resource(nvc0->idxbuf.buffer),
- nvc0->idxbuf.offset, NOUVEAU_BO_RD);
- if (!data)
- return;
+ const void *data = nvc0->idxbuf.user_buffer;
while (instance_count--) {
PUSH_SPACE(push, 2);
struct nvc0_context *nvc0 = nvc0_context(pipe);
struct nouveau_pushbuf *push = nvc0->base.pushbuf;
+ /* NOTE: caller must ensure that (min_index + index_bias) is >= 0 */
+ nvc0->vb_elt_first = info->min_index + info->index_bias;
+ nvc0->vb_elt_limit = info->max_index - info->min_index;
+ nvc0->instance_off = info->start_instance;
+ nvc0->instance_max = info->instance_count - 1;
+
/* For picking only a few vertices from a large user buffer, push is better,
* if index count is larger and we expect repeated vertices, suggest upload.
*/
- nvc0->vbo_push_hint = /* the 64 is heuristic */
- !(info->indexed &&
- ((info->max_index - info->min_index + 64) < info->count));
-
- nvc0->vbo_min_index = info->min_index;
- nvc0->vbo_max_index = info->max_index;
-
- if (nvc0->vbo_push_hint != !!nvc0->vbo_fifo)
- nvc0->dirty |= NVC0_NEW_ARRAYS;
-
- if (nvc0->vbo_user && !(nvc0->dirty & (NVC0_NEW_VERTEX | NVC0_NEW_ARRAYS)))
- nvc0_update_user_vbufs(nvc0);
+ nvc0->vbo_push_hint =
+ info->indexed && (nvc0->vb_elt_limit >= (info->count * 2));
+
+ /* Check whether we want to switch vertex-submission mode. */
+ if (nvc0->vbo_user && !(nvc0->dirty & (NVC0_NEW_ARRAYS | NVC0_NEW_VERTEX))) {
+ if (nvc0->vbo_push_hint != !!nvc0->state.vbo_mode)
+ if (nvc0->state.vbo_mode != 3)
+ nvc0->dirty |= NVC0_NEW_ARRAYS;
+
+ if (!(nvc0->dirty & NVC0_NEW_ARRAYS) && nvc0->state.vbo_mode == 0) {
+ if (nvc0->vertex->shared_slots)
+ nvc0_update_user_vbufs_shared(nvc0);
+ else
+ nvc0_update_user_vbufs(nvc0);
+ }
+ }
/* 8 as minimum to avoid immediate double validation of new buffers */
nvc0_state_validate(nvc0, ~0, 8);
push->kick_notify = nvc0_draw_vbo_kick_notify;
- if (nvc0->vbo_fifo) {
+ if (nvc0->state.vbo_mode) {
nvc0_push_vbo(nvc0, info);
push->kick_notify = nvc0_default_kick_notify;
return;
}
if (nvc0->base.vbo_dirty) {
- BEGIN_NVC0(push, NVC0_3D(VERTEX_ARRAY_FLUSH), 1);
- PUSH_DATA (push, 0);
+ IMMED_NVC0(push, NVC0_3D(VERTEX_ARRAY_FLUSH), 0);
nvc0->base.vbo_dirty = FALSE;
}
- if (unlikely(info->count_from_stream_output)) {
- nvc0_draw_stream_output(nvc0, info);
- } else
- if (!info->indexed) {
- nvc0_draw_arrays(nvc0,
- info->mode, info->start, info->count,
- info->instance_count);
- } else {
+ if (info->indexed) {
boolean shorten = info->max_index <= 65535;
- assert(nvc0->idxbuf.buffer);
-
if (info->primitive_restart != nvc0->state.prim_restart) {
if (info->primitive_restart) {
BEGIN_NVC0(push, NVC0_3D(PRIM_RESTART_ENABLE), 2);
nvc0_draw_elements(nvc0, shorten,
info->mode, info->start, info->count,
info->instance_count, info->index_bias);
+ } else
+ if (unlikely(info->count_from_stream_output)) {
+ nvc0_draw_stream_output(nvc0, info);
+ } else {
+ nvc0_draw_arrays(nvc0,
+ info->mode, info->start, info->count,
+ info->instance_count);
}
push->kick_notify = nvc0_default_kick_notify;
projects / mesa.git / blobdiff