void _tnl_register_fastpath( struct tnl_clipspace *vtx,
GLboolean match_strides )
{
struct tnl_clipspace_fastpath *fastpath = CALLOC_STRUCT(tnl_clipspace_fastpath);
GLuint i;
if (fastpath == NULL) {
_mesa_error_no_memory(__func__);
return;
}
fastpath->vertex_size = vtx->vertex_size;
fastpath->attr_count = vtx->attr_count;
fastpath->match_strides = match_strides;
fastpath->func = vtx->emit;
fastpath->attr = malloc(vtx->attr_count * sizeof(fastpath->attr[0]));
if (fastpath->attr == NULL) {
free(fastpath);
_mesa_error_no_memory(__func__);
return;
}
for (i = 0; i < vtx->attr_count; i++) {
fastpath->attr[i].format = vtx->attr[i].format;
fastpath->attr[i].stride = vtx->attr[i].inputstride;
fastpath->attr[i].size = vtx->attr[i].inputsize;
fastpath->attr[i].offset = vtx->attr[i].vertoffset;
}
fastpath->next = vtx->fastpath;
vtx->fastpath = fastpath;
}
/**
* The NewBufferObject() driver hook.
*
* Allocates a new intel_buffer_object structure and initializes it.
*
* There is some duplication between mesa's bufferobjects and our
* bufmgr buffers. Both have an integer handle and a hashtable to
* lookup an opaque structure. It would be nice if the handles and
* internal structure where somehow shared.
*/
static struct gl_buffer_object *
brw_new_buffer_object(struct gl_context * ctx, GLuint name)
{
struct intel_buffer_object *obj = CALLOC_STRUCT(intel_buffer_object);
if (!obj) {
_mesa_error_no_memory(__func__);
}
_mesa_initialize_buffer_object(ctx, &obj->Base, name);
obj->buffer = NULL;
return &obj->Base;
}
static GLintptr
register_surface(struct gl_context *ctx, GLboolean isOutput,
const GLvoid *vdpSurface, GLenum target,
GLsizei numTextureNames, const GLuint *textureNames)
{
struct vdp_surface *surf;
int i;
if (!ctx->vdpDevice || !ctx->vdpGetProcAddress || !ctx->vdpSurfaces) {
_mesa_error(ctx, GL_INVALID_OPERATION, "VDPAURegisterSurfaceNV");
return (GLintptr)NULL;
}
if (target != GL_TEXTURE_2D && target != GL_TEXTURE_RECTANGLE) {
_mesa_error(ctx, GL_INVALID_ENUM, "VDPAURegisterSurfaceNV");
return (GLintptr)NULL;
}
if (target == GL_TEXTURE_RECTANGLE && !ctx->Extensions.NV_texture_rectangle) {
_mesa_error(ctx, GL_INVALID_ENUM, "VDPAURegisterSurfaceNV");
return (GLintptr)NULL;
}
surf = CALLOC_STRUCT( vdp_surface );
if (surf == NULL) {
_mesa_error_no_memory("VDPAURegisterSurfaceNV");
return (GLintptr)NULL;
}
surf->vdpSurface = vdpSurface;
surf->target = target;
surf->access = GL_READ_WRITE;
surf->state = GL_SURFACE_REGISTERED_NV;
surf->output = isOutput;
for (i = 0; i < numTextureNames; ++i) {
struct gl_texture_object *tex;
tex = _mesa_lookup_texture(ctx, textureNames[i]);
if (tex == NULL) {
free(surf);
_mesa_error(ctx, GL_INVALID_OPERATION,
"VDPAURegisterSurfaceNV(texture ID not found)");
return (GLintptr)NULL;
}
_mesa_lock_texture(ctx, tex);
if (tex->Immutable) {
_mesa_unlock_texture(ctx, tex);
free(surf);
_mesa_error(ctx, GL_INVALID_OPERATION,
"VDPAURegisterSurfaceNV(texture is immutable)");
return (GLintptr)NULL;
}
if (tex->Target == 0)
tex->Target = target;
else if (tex->Target != target) {
_mesa_unlock_texture(ctx, tex);
free(surf);
_mesa_error(ctx, GL_INVALID_OPERATION,
"VDPAURegisterSurfaceNV(target mismatch)");
return (GLintptr)NULL;
}
/* This will disallow respecifying the storage. */
tex->Immutable = GL_TRUE;
_mesa_unlock_texture(ctx, tex);
_mesa_reference_texobj(&surf->textures[i], tex);
}
_mesa_set_add(ctx->vdpSurfaces, surf);
return (GLintptr)surf;
}
void
link_uniform_blocks(void *mem_ctx,
struct gl_context *ctx,
struct gl_shader_program *prog,
struct gl_linked_shader *shader,
struct gl_uniform_block **ubo_blocks,
unsigned *num_ubo_blocks,
struct gl_uniform_block **ssbo_blocks,
unsigned *num_ssbo_blocks)
{
/* This hash table will track all of the uniform blocks that have been
* encountered. Since blocks with the same block-name must be the same,
* the hash is organized by block-name.
*/
struct hash_table *block_hash =
_mesa_hash_table_create(mem_ctx, _mesa_key_hash_string,
_mesa_key_string_equal);
if (block_hash == NULL) {
_mesa_error_no_memory(__func__);
linker_error(prog, "out of memory\n");
return;
}
/* Determine which uniform blocks are active.
*/
link_uniform_block_active_visitor v(mem_ctx, block_hash, prog);
visit_list_elements(&v, shader->ir);
/* Count the number of active uniform blocks. Count the total number of
* active slots in those uniform blocks.
*/
unsigned num_ubo_variables = 0;
unsigned num_ssbo_variables = 0;
count_block_size block_size;
struct hash_entry *entry;
hash_table_foreach (block_hash, entry) {
struct link_uniform_block_active *const b =
(struct link_uniform_block_active *) entry->data;
assert((b->array != NULL) == b->type->is_array());
if (b->array != NULL &&
(b->type->without_array()->interface_packing ==
GLSL_INTERFACE_PACKING_PACKED)) {
b->type = resize_block_array(b->type, b->array);
b->var->type = b->type;
}
block_size.num_active_uniforms = 0;
block_size.process(b->type->without_array(), "");
if (b->array != NULL) {
unsigned aoa_size = b->type->arrays_of_arrays_size();
if (b->is_shader_storage) {
*num_ssbo_blocks += aoa_size;
num_ssbo_variables += aoa_size * block_size.num_active_uniforms;
} else {
*num_ubo_blocks += aoa_size;
num_ubo_variables += aoa_size * block_size.num_active_uniforms;
}
} else {
if (b->is_shader_storage) {
(*num_ssbo_blocks)++;
num_ssbo_variables += block_size.num_active_uniforms;
} else {
(*num_ubo_blocks)++;
num_ubo_variables += block_size.num_active_uniforms;
}
}
}
create_buffer_blocks(mem_ctx, ctx, prog, ubo_blocks, *num_ubo_blocks,
block_hash, num_ubo_variables, true);
create_buffer_blocks(mem_ctx, ctx, prog, ssbo_blocks, *num_ssbo_blocks,
block_hash, num_ssbo_variables, false);
_mesa_hash_table_destroy(block_hash, NULL);
}
/* Adjust primitives, indices and vertex definitions so that min_index
* becomes zero. There are lots of reasons for wanting to do this, eg:
*
* Software tnl:
* - any time min_index != 0, otherwise unused vertices lower than
* min_index will be transformed.
*
* Hardware tnl:
* - if ib != NULL and min_index != 0, otherwise vertices lower than
* min_index will be uploaded. Requires adjusting index values.
*
* - if ib == NULL and min_index != 0, just for convenience so this doesn't
* have to be handled within the driver.
*
* Hardware tnl with VBO support:
* - as above, but only when vertices are not (all?) in VBO's.
* - can't save time by trying to upload half a vbo - typically it is
* all or nothing.
*/
void vbo_rebase_prims( struct gl_context *ctx,
const struct gl_vertex_array *arrays[],
const struct _mesa_prim *prim,
GLuint nr_prims,
const struct _mesa_index_buffer *ib,
GLuint min_index,
GLuint max_index,
vbo_draw_func draw )
{
struct gl_vertex_array tmp_arrays[VERT_ATTRIB_MAX];
const struct gl_vertex_array *tmp_array_pointers[VERT_ATTRIB_MAX];
struct _mesa_index_buffer tmp_ib;
struct _mesa_prim *tmp_prims = NULL;
const struct gl_vertex_array **saved_arrays = ctx->Array._DrawArrays;
void *tmp_indices = NULL;
GLuint i;
assert(min_index != 0);
if (0)
printf("%s %d..%d\n", __func__, min_index, max_index);
/* XXX this path is disabled for now.
* There's rendering corruption in some apps when it's enabled.
*/
if (0 && ib && ctx->Extensions.ARB_draw_elements_base_vertex) {
/* If we can just tell the hardware or the TNL to interpret our
* indices with a different base, do so.
*/
tmp_prims = malloc(sizeof(*prim) * nr_prims);
if (tmp_prims == NULL) {
_mesa_error_no_memory(__func__);
return;
}
for (i = 0; i < nr_prims; i++) {
tmp_prims[i] = prim[i];
tmp_prims[i].basevertex -= min_index;
}
prim = tmp_prims;
} else if (ib) {
/* Unfortunately need to adjust each index individually.
*/
GLboolean map_ib = ib->obj->Name &&
!ib->obj->Mappings[MAP_INTERNAL].Pointer;
void *ptr;
if (map_ib)
ctx->Driver.MapBufferRange(ctx, 0, ib->obj->Size, GL_MAP_READ_BIT,
ib->obj, MAP_INTERNAL);
ptr = ADD_POINTERS(ib->obj->Mappings[MAP_INTERNAL].Pointer, ib->ptr);
/* Some users might prefer it if we translated elements to
* GLuints here. Others wouldn't...
*/
switch (ib->type) {
case GL_UNSIGNED_INT:
tmp_indices = rebase_GLuint( ptr, ib->count, min_index );
break;
case GL_UNSIGNED_SHORT:
tmp_indices = rebase_GLushort( ptr, ib->count, min_index );
break;
case GL_UNSIGNED_BYTE:
tmp_indices = rebase_GLubyte( ptr, ib->count, min_index );
break;
}
if (map_ib)
ctx->Driver.UnmapBuffer(ctx, ib->obj, MAP_INTERNAL);
if (tmp_indices == NULL) {
return;
}
tmp_ib.obj = ctx->Shared->NullBufferObj;
tmp_ib.ptr = tmp_indices;
tmp_ib.count = ib->count;
tmp_ib.type = ib->type;
ib = &tmp_ib;
}
else {
/* Otherwise the primitives need adjustment.
*/
tmp_prims = malloc(sizeof(*prim) * nr_prims);
if (tmp_prims == NULL) {
_mesa_error_no_memory(__func__);
return;
}
for (i = 0; i < nr_prims; i++) {
/* If this fails, it could indicate an application error:
*/
assert(prim[i].start >= min_index);
tmp_prims[i] = prim[i];
tmp_prims[i].start -= min_index;
}
prim = tmp_prims;
}
/* Just need to adjust the pointer values on each incoming array.
* This works for VBO and non-vbo rendering and shouldn't pesimize
* VBO-based upload schemes. However this may still not be a fast
* path for hardware tnl for VBO based rendering as most machines
* will be happier if you just specify a starting vertex value in
* each primitive.
*
* For drivers with hardware tnl, you only want to do this if you
* are forced to, eg non-VBO indexed rendering with start != 0.
*/
for (i = 0; i < VERT_ATTRIB_MAX; i++) {
tmp_arrays[i] = *arrays[i];
tmp_arrays[i].Ptr += min_index * tmp_arrays[i].StrideB;
tmp_array_pointers[i] = &tmp_arrays[i];
}
/* Re-issue the draw call.
*/
ctx->Array._DrawArrays = tmp_array_pointers;
ctx->NewDriverState |= ctx->DriverFlags.NewArray;
draw( ctx,
prim,
nr_prims,
ib,
GL_TRUE,
0,
max_index - min_index,
NULL, 0, NULL );
ctx->Array._DrawArrays = saved_arrays;
ctx->NewDriverState |= ctx->DriverFlags.NewArray;
free(tmp_indices);
free(tmp_prims);
}
