From: Marek Olšák <marek.ol...@amd.com>
This is the best place to do it. Now drivers without u_vbuf don't have to
do it.
---
src/mesa/state_tracker/st_atom_array.c | 56 ++++++++++++++++++++++++----------
src/mesa/state_tracker/st_context.c | 2 ++
src/mesa/state_tracker/st_context.h | 1 +
3 files changed, 43 insertions(+), 16 deletions(-)
diff --git a/src/mesa/state_tracker/st_atom_array.c
b/src/mesa/state_tracker/st_atom_array.c
index cc9cac1..813468b 100644
--- a/src/mesa/state_tracker/st_atom_array.c
+++ b/src/mesa/state_tracker/st_atom_array.c
@@ -37,20 +37,21 @@
*/
#include "st_context.h"
#include "st_atom.h"
#include "st_cb_bufferobjects.h"
#include "st_draw.h"
#include "st_program.h"
#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"
/* vertex_formats[gltype - GL_BYTE][integer*2 + normalized][size - 1] */
static const uint16_t vertex_formats[][4][4] = {
{ /* GL_BYTE */
{
PIPE_FORMAT_R8_SSCALED,
PIPE_FORMAT_R8G8_SSCALED,
PIPE_FORMAT_R8G8B8_SSCALED,
@@ -327,20 +328,25 @@ is_interleaved_arrays(const struct st_vertex_program *vp,
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(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 */
firstStride = stride;
firstPtr = array->Ptr;
firstBufObj = bufObj;
userSpaceBuffer = !bufObj || !bufObj->Name;
}
else {
/* check if other arrays interleave with the first, in same buffer */
@@ -564,20 +570,21 @@ setup_interleaved_attribs(struct st_context *st,
static void
setup_non_interleaved_attribs(struct st_context *st,
const struct st_vertex_program *vp,
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 < 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;
@@ -601,54 +608,71 @@ setup_non_interleaved_attribs(struct st_context *st,
if (!stobj || !stobj->buffer) {
st->vertex_array_out_of_memory = true;
return; /* out-of-memory error probably */
}
vbuffer[bufidx].buffer.resource = stobj->buffer;
vbuffer[bufidx].is_user_buffer = false;
vbuffer[bufidx].buffer_offset = pointer_to_offset(array->Ptr);
}
else {
- /* wrap user data */
- void *ptr;
-
- if (array->Ptr) {
- ptr = (void *) array->Ptr;
- }
- else {
- /* no array, use ctx->Current.Attrib[] value */
- ptr = (void *) ctx->Current.Attrib[mesaAttr];
- stride = 0;
+ if (stride == 0) {
+ 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, 32, 32, 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;
}
-
- 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 */
vbuffer[bufidx].stride = stride; /* in bytes */
src_format = st_pipe_vertex_format(array->Type,
array->Size,
array->Format,
array->Normalized,
array->Integer);
init_velement_lowered(vp, velements, 0, src_format,
array->InstanceDivisor, bufidx,
array->Size, array->Doubles, &attr);
}
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;
unsigned num_inputs;
st->vertex_array_out_of_memory = FALSE;
diff --git a/src/mesa/state_tracker/st_context.c
b/src/mesa/state_tracker/st_context.c
index d997911..c901764 100644
--- a/src/mesa/state_tracker/st_context.c
+++ b/src/mesa/state_tracker/st_context.c
@@ -331,20 +331,22 @@ st_create_context_priv( struct gl_context *ctx, struct
pipe_context *pipe,
st->ctx = ctx;
st->pipe = pipe;
/* state tracker needs the VBO module */
_vbo_CreateContext(ctx);
st->dirty = ST_ALL_STATES_MASK;
st->has_user_constbuf =
screen->get_param(screen, PIPE_CAP_USER_CONSTANT_BUFFERS);
+ st->can_bind_const_buffer_as_vertex =
+ screen->get_param(screen, PIPE_CAP_CAN_BIND_CONST_BUFFER_AS_VERTEX);
/* Drivers still have to upload zero-stride vertex attribs manually
* with the GL core profile, but they don't have to deal with any complex
* user vertex buffer uploads.
*/
unsigned vbuf_flags =
ctx->API == API_OPENGL_CORE ? U_VBUF_FLAG_NO_USER_VBOS : 0;
st->cso_context = cso_create_context(pipe, vbuf_flags);
st_init_atoms( st );
diff --git a/src/mesa/state_tracker/st_context.h
b/src/mesa/state_tracker/st_context.h
index eb1a163..68fd9b9 100644
--- a/src/mesa/state_tracker/st_context.h
+++ b/src/mesa/state_tracker/st_context.h
@@ -97,20 +97,21 @@ struct st_context
boolean has_time_elapsed;
boolean has_shader_model3;
boolean has_etc1;
boolean has_etc2;
boolean prefer_blit_based_texture_transfer;
boolean force_persample_in_shader;
boolean has_shareable_shaders;
boolean has_half_float_packing;
boolean has_multi_draw_indirect;
boolean has_user_constbuf;
+ boolean can_bind_const_buffer_as_vertex;
/**
* If a shader can be created when we get its source.
* This means it has only 1 variant, not counting glBitmap and
* glDrawPixels.
*/
boolean shader_has_one_variant[MESA_SHADER_STAGES];
boolean needs_texcoord_semantic;
boolean apply_texture_swizzle_to_border_color;
--
2.7.4
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
