#include "cso_cache/cso_context.h"
#include "util/u_math.h"
+#include "util/u_upload_mgr.h"
#include "main/bufferobj.h"
#include "main/glformats.h"
}
static const struct gl_vertex_array *
-get_client_array(const struct st_vertex_program *vp,
- const struct gl_vertex_array **arrays,
- int attr)
+get_client_array(const struct gl_vertex_array **arrays,
+ unsigned mesaAttr)
{
- const GLuint mesaAttr = vp->index_to_input[attr];
/* st_program uses 0xffffffff to denote a double placeholder attribute */
if (mesaAttr == ST_DOUBLE_ATTRIB_PLACEHOLDER)
return NULL;
*/
static GLboolean
is_interleaved_arrays(const struct st_vertex_program *vp,
- const struct st_vp_variant *vpv,
- const struct gl_vertex_array **arrays)
+ const struct gl_vertex_array **arrays,
+ unsigned num_inputs)
{
GLuint attr;
const struct gl_buffer_object *firstBufObj = NULL;
const GLubyte *firstPtr = NULL;
GLboolean userSpaceBuffer = GL_FALSE;
- for (attr = 0; attr < vpv->num_inputs; attr++) {
+ for (attr = 0; attr < num_inputs; attr++) {
const struct gl_vertex_array *array;
const struct gl_buffer_object *bufObj;
GLsizei stride;
- array = get_client_array(vp, arrays, attr);
+ array = get_client_array(arrays, vp->index_to_input[attr]);
if (!array)
continue;
stride = array->StrideB; /* in bytes */
+
+ /* To keep things simple, don't allow interleaved zero-stride attribs. */
+ if (stride == 0)
+ return false;
+
bufObj = array->BufferObj;
if (attr == 0) {
/* save info about the first array */
* \param vbuffer returns vertex buffer info
* \param velements returns vertex element info
*/
-static boolean
+static void
setup_interleaved_attribs(struct st_context *st,
const struct st_vertex_program *vp,
- const struct st_vp_variant *vpv,
- const struct gl_vertex_array **arrays)
+ const struct gl_vertex_array **arrays,
+ unsigned num_inputs)
{
struct pipe_vertex_buffer vbuffer;
struct pipe_vertex_element velements[PIPE_MAX_ATTRIBS] = {{0}};
/* Find the lowest address of the arrays we're drawing,
* Init bufobj and stride.
*/
- if (vpv->num_inputs) {
+ if (num_inputs) {
const struct gl_vertex_array *array;
- array = get_client_array(vp, arrays, 0);
+ array = get_client_array(arrays, vp->index_to_input[0]);
assert(array);
/* Since we're doing interleaved arrays, we know there'll be at most
low_addr = arrays[vp->index_to_input[0]]->Ptr;
- for (attr = 1; attr < vpv->num_inputs; attr++) {
+ for (attr = 1; attr < num_inputs; attr++) {
const GLubyte *start;
- array = get_client_array(vp, arrays, attr);
+ array = get_client_array(arrays, vp->index_to_input[attr]);
if (!array)
continue;
start = array->Ptr;
/* are the arrays in user space? */
usingVBO = _mesa_is_bufferobj(bufobj);
- for (attr = 0; attr < vpv->num_inputs;) {
+ for (attr = 0; attr < num_inputs;) {
const struct gl_vertex_array *array;
unsigned src_offset;
unsigned src_format;
- array = get_client_array(vp, arrays, attr);
+ array = get_client_array(arrays, vp->index_to_input[attr]);
assert(array);
src_offset = (unsigned) (array->Ptr - low_addr);
/*
* Return the vbuffer info and setup user-space attrib info, if needed.
*/
- if (vpv->num_inputs == 0) {
+ if (num_inputs == 0) {
/* just defensive coding here */
vbuffer.buffer.resource = NULL;
vbuffer.is_user_buffer = false;
struct st_buffer_object *stobj = st_buffer_object(bufobj);
if (!stobj || !stobj->buffer) {
- return FALSE; /* out-of-memory error probably */
+ st->vertex_array_out_of_memory = true;
+ return; /* out-of-memory error probably */
}
vbuffer.buffer.resource = stobj->buffer;
vbuffer.is_user_buffer = !!low_addr; /* if NULL, then unbind */
vbuffer.buffer_offset = 0;
vbuffer.stride = stride;
+
+ if (low_addr)
+ st->draw_needs_minmax_index = true;
}
- set_vertex_attribs(st, &vbuffer, vpv->num_inputs ? 1 : 0,
- velements, vpv->num_inputs);
- return TRUE;
+ set_vertex_attribs(st, &vbuffer, num_inputs ? 1 : 0,
+ velements, num_inputs);
}
/**
* \param vbuffer returns vertex buffer info
* \param velements returns vertex element info
*/
-static boolean
+static void
setup_non_interleaved_attribs(struct st_context *st,
const struct st_vertex_program *vp,
- const struct st_vp_variant *vpv,
- const struct gl_vertex_array **arrays)
+ const struct gl_vertex_array **arrays,
+ unsigned num_inputs)
{
struct gl_context *ctx = st->ctx;
struct pipe_vertex_buffer vbuffer[PIPE_MAX_ATTRIBS];
struct pipe_vertex_element velements[PIPE_MAX_ATTRIBS] = {{0}};
unsigned num_vbuffers = 0;
+ unsigned unref_buffers = 0;
GLuint attr;
- for (attr = 0; attr < vpv->num_inputs;) {
- const GLuint mesaAttr = vp->index_to_input[attr];
+ for (attr = 0; attr < num_inputs;) {
+ const unsigned mesaAttr = vp->index_to_input[attr];
const struct gl_vertex_array *array;
struct gl_buffer_object *bufobj;
GLsizei stride;
unsigned src_format;
unsigned bufidx;
- array = get_client_array(vp, arrays, attr);
+ array = get_client_array(arrays, mesaAttr);
assert(array);
bufidx = num_vbuffers++;
struct st_buffer_object *stobj = st_buffer_object(bufobj);
if (!stobj || !stobj->buffer) {
- return FALSE; /* out-of-memory error probably */
+ st->vertex_array_out_of_memory = true;
+ return; /* out-of-memory error probably */
}
vbuffer[bufidx].buffer.resource = stobj->buffer;
vbuffer[bufidx].buffer_offset = pointer_to_offset(array->Ptr);
}
else {
- /* wrap user data */
- void *ptr;
+ if (stride == 0) {
+ unsigned size = array->_ElementSize;
+ /* This is optimal for GPU cache line usage if the upload size
+ * is <= cache line size.
+ */
+ unsigned alignment = util_next_power_of_two(size);
+ void *ptr = array->Ptr ? (void*)array->Ptr :
+ (void*)ctx->Current.Attrib[mesaAttr];
+
+ vbuffer[bufidx].is_user_buffer = false;
+ vbuffer[bufidx].buffer.resource = NULL;
+
+ /* Use const_uploader for zero-stride vertex attributes, because
+ * it may use a better memory placement than stream_uploader.
+ * The reason is that zero-stride attributes can be fetched many
+ * times (thousands of times), so a better placement is going to
+ * perform better.
+ *
+ * Upload the maximum possible size, which is 4x GLdouble = 32.
+ */
+ u_upload_data(st->can_bind_const_buffer_as_vertex ?
+ st->pipe->const_uploader :
+ st->pipe->stream_uploader,
+ 0, size, alignment, ptr,
+ &vbuffer[bufidx].buffer_offset,
+ &vbuffer[bufidx].buffer.resource);
+ unref_buffers |= 1u << bufidx;
+ } else {
+ assert(array->Ptr);
+ vbuffer[bufidx].buffer.user = array->Ptr;
+ vbuffer[bufidx].is_user_buffer = true;
+ vbuffer[bufidx].buffer_offset = 0;
- if (array->Ptr) {
- ptr = (void *) array->Ptr;
+ if (!array->InstanceDivisor)
+ st->draw_needs_minmax_index = true;
}
- else {
- /* no array, use ctx->Current.Attrib[] value */
- ptr = (void *) ctx->Current.Attrib[mesaAttr];
- stride = 0;
- }
-
- assert(ptr);
-
- vbuffer[bufidx].buffer.user = ptr;
- vbuffer[bufidx].is_user_buffer = !!ptr; /* if NULL, then unbind */
- vbuffer[bufidx].buffer_offset = 0;
}
/* common-case setup */
array->Size, array->Doubles, &attr);
}
- set_vertex_attribs(st, vbuffer, num_vbuffers, velements, vpv->num_inputs);
- return TRUE;
+ if (!ctx->Const.AllowMappedBuffersDuringExecution) {
+ u_upload_unmap(st->pipe->stream_uploader);
+ }
+
+ set_vertex_attribs(st, vbuffer, num_vbuffers, velements, num_inputs);
+
+ /* Unreference uploaded zero-stride vertex buffers. */
+ while (unref_buffers) {
+ unsigned i = u_bit_scan(&unref_buffers);
+ pipe_resource_reference(&vbuffer[i].buffer.resource, NULL);
+ }
}
void st_update_array(struct st_context *st)
struct gl_context *ctx = st->ctx;
const struct gl_vertex_array **arrays = ctx->Array._DrawArrays;
const struct st_vertex_program *vp;
- const struct st_vp_variant *vpv;
+ unsigned num_inputs;
st->vertex_array_out_of_memory = FALSE;
+ st->draw_needs_minmax_index = false;
/* No drawing has been done yet, so do nothing. */
if (!arrays)
/* vertex program validation must be done before this */
vp = st->vp;
- vpv = st->vp_variant;
+ num_inputs = st->vp_variant->num_inputs;
- if (is_interleaved_arrays(vp, vpv, arrays)) {
- if (!setup_interleaved_attribs(st, vp, vpv, arrays)) {
- st->vertex_array_out_of_memory = TRUE;
- return;
- }
- }
- else {
- if (!setup_non_interleaved_attribs(st, vp, vpv, arrays)) {
- st->vertex_array_out_of_memory = TRUE;
- return;
- }
- }
+ if (is_interleaved_arrays(vp, arrays, num_inputs))
+ setup_interleaved_attribs(st, vp, arrays, num_inputs);
+ else
+ setup_non_interleaved_attribs(st, vp, arrays, num_inputs);
}
projects / mesa.git / blobdiff