static void
copy_array_to_vbo_array(struct brw_context *brw,
struct brw_vertex_element *element,
int min, int max,
struct brw_vertex_buffer *buffer,
GLuint dst_stride)
{
int src_stride = element->glarray->StrideB;
const unsigned char *src = element->glarray->Ptr + min * src_stride;
int count = max - min + 1;
GLuint size = count * dst_stride;
if (dst_stride == src_stride) {
intel_upload_data(&brw->intel, src, size, dst_stride,
&buffer->bo, &buffer->offset);
} else {
char * const map = intel_upload_map(&brw->intel, size, dst_stride);
char *dst = map;
while (count--) {
memcpy(dst, src, dst_stride);
src += src_stride;
dst += dst_stride;
}
intel_upload_unmap(&brw->intel, map, size, dst_stride,
&buffer->bo, &buffer->offset);
}
buffer->stride = dst_stride;
}
static void
copy_array_to_vbo_array(struct brw_context *brw,
struct brw_vertex_element *element,
int min, int max,
struct brw_vertex_buffer *buffer,
GLuint dst_stride)
{
if (min == -1) {
/* If we don't have computed min/max bounds, then this must be a use of
* the current attribute, which has a 0 stride. Otherwise, we wouldn't
* know what data to upload.
*/
assert(element->glarray->StrideB == 0);
intel_upload_data(&brw->intel, element->glarray->Ptr,
element->element_size,
element->element_size,
&buffer->bo, &buffer->offset);
buffer->stride = 0;
return;
}
int src_stride = element->glarray->StrideB;
const unsigned char *src = element->glarray->Ptr + min * src_stride;
int count = max - min + 1;
GLuint size = count * dst_stride;
if (dst_stride == src_stride) {
intel_upload_data(&brw->intel, src, size, dst_stride,
&buffer->bo, &buffer->offset);
} else {
char * const map = intel_upload_map(&brw->intel, size, dst_stride);
char *dst = map;
while (count--) {
memcpy(dst, src, dst_stride);
src += src_stride;
dst += dst_stride;
}
intel_upload_unmap(&brw->intel, map, size, dst_stride,
&buffer->bo, &buffer->offset);
}
buffer->stride = dst_stride;
}
static void
copy_array_to_vbo_array(struct brw_context *brw,
struct brw_vertex_element *element,
int min, int max,
struct brw_vertex_buffer *buffer,
GLuint dst_stride)
{
const int src_stride = element->glarray->StrideB;
/* If the source stride is zero, we just want to upload the current
* attribute once and set the buffer's stride to 0. There's no need
* to replicate it out.
*/
if (src_stride == 0) {
intel_upload_data(brw, element->glarray->Ptr,
element->glarray->_ElementSize,
element->glarray->_ElementSize,
&buffer->bo, &buffer->offset);
buffer->stride = 0;
return;
}
const unsigned char *src = element->glarray->Ptr + min * src_stride;
int count = max - min + 1;
GLuint size = count * dst_stride;
if (dst_stride == src_stride) {
intel_upload_data(brw, src, size, dst_stride,
&buffer->bo, &buffer->offset);
} else {
char * const map = intel_upload_map(brw, size, dst_stride);
char *dst = map;
while (count--) {
memcpy(dst, src, dst_stride);
src += src_stride;
dst += dst_stride;
}
intel_upload_unmap(brw, map, size, dst_stride,
&buffer->bo, &buffer->offset);
}
buffer->stride = dst_stride;
}
static void brw_prepare_vertices(struct brw_context *brw)
{
struct gl_context *ctx = &brw->intel.ctx;
struct intel_context *intel = intel_context(ctx);
/* CACHE_NEW_VS_PROG */
GLbitfield64 vs_inputs = brw->vs.prog_data->inputs_read;
const unsigned char *ptr = NULL;
GLuint interleaved = 0, total_size = 0;
unsigned int min_index = brw->vb.min_index;
unsigned int max_index = brw->vb.max_index;
int delta, i, j;
GLboolean can_merge_uploads = GL_TRUE;
struct brw_vertex_element *upload[VERT_ATTRIB_MAX];
GLuint nr_uploads = 0;
/* First build an array of pointers to ve's in vb.inputs_read
*/
if (0)
printf("%s %d..%d\n", __FUNCTION__, min_index, max_index);
/* Accumulate the list of enabled arrays. */
brw->vb.nr_enabled = 0;
while (vs_inputs) {
GLuint i = ffsll(vs_inputs) - 1;
struct brw_vertex_element *input = &brw->vb.inputs[i];
vs_inputs &= ~BITFIELD64_BIT(i);
if (input->glarray->Size && get_size(input->glarray->Type))
brw->vb.enabled[brw->vb.nr_enabled++] = input;
}
if (brw->vb.nr_enabled == 0)
return;
if (brw->vb.nr_buffers)
goto prepare;
for (i = j = 0; i < brw->vb.nr_enabled; i++) {
struct brw_vertex_element *input = brw->vb.enabled[i];
const struct gl_client_array *glarray = input->glarray;
int type_size = get_size(glarray->Type);
input->element_size = type_size * glarray->Size;
if (_mesa_is_bufferobj(glarray->BufferObj)) {
struct intel_buffer_object *intel_buffer =
intel_buffer_object(glarray->BufferObj);
int k;
for (k = 0; k < i; k++) {
const struct gl_client_array *other = brw->vb.enabled[k]->glarray;
if (glarray->BufferObj == other->BufferObj &&
glarray->StrideB == other->StrideB &&
glarray->InstanceDivisor == other->InstanceDivisor &&
(uintptr_t)(glarray->Ptr - other->Ptr) < glarray->StrideB)
{
input->buffer = brw->vb.enabled[k]->buffer;
input->offset = glarray->Ptr - other->Ptr;
break;
}
}
if (k == i) {
struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
/* Named buffer object: Just reference its contents directly. */
buffer->bo = intel_bufferobj_source(intel,
intel_buffer, type_size,
&buffer->offset);
drm_intel_bo_reference(buffer->bo);
buffer->offset += (uintptr_t)glarray->Ptr;
buffer->stride = glarray->StrideB;
buffer->step_rate = glarray->InstanceDivisor;
input->buffer = j++;
input->offset = 0;
}
/* This is a common place to reach if the user mistakenly supplies
* a pointer in place of a VBO offset. If we just let it go through,
* we may end up dereferencing a pointer beyond the bounds of the
* GTT. We would hope that the VBO's max_index would save us, but
* Mesa appears to hand us min/max values not clipped to the
* array object's _MaxElement, and _MaxElement frequently appears
* to be wrong anyway.
*
* The VBO spec allows application termination in this case, and it's
* probably a service to the poor programmer to do so rather than
* trying to just not render.
*/
assert(input->offset < brw->vb.buffers[input->buffer].bo->size);
} else {
/* Queue the buffer object up to be uploaded in the next pass,
* when we've decided if we're doing interleaved or not.
*/
if (nr_uploads == 0) {
/* Position array not properly enabled:
*/
if (input->attrib == VERT_ATTRIB_POS && glarray->StrideB == 0) {
intel->Fallback = true; /* boolean, not bitfield */
return;
}
interleaved = glarray->StrideB;
ptr = glarray->Ptr;
}
else if (interleaved != glarray->StrideB ||
(uintptr_t)(glarray->Ptr - ptr) > interleaved)
{
interleaved = 0;
}
else if ((uintptr_t)(glarray->Ptr - ptr) & (type_size -1))
{
/* enforce natural alignment (for doubles) */
interleaved = 0;
}
upload[nr_uploads++] = input;
total_size = ALIGN(total_size, type_size);
total_size += input->element_size;
if (glarray->InstanceDivisor != 0) {
can_merge_uploads = GL_FALSE;
}
}
}
/* If we need to upload all the arrays, then we can trim those arrays to
* only the used elements [min_index, max_index] so long as we adjust all
* the values used in the 3DPRIMITIVE i.e. by setting the vertex bias.
*/
brw->vb.start_vertex_bias = 0;
delta = min_index;
if (nr_uploads == brw->vb.nr_enabled) {
brw->vb.start_vertex_bias = -delta;
delta = 0;
}
if (delta && !brw->intel.intelScreen->relaxed_relocations)
min_index = delta = 0;
/* Handle any arrays to be uploaded. */
if (nr_uploads > 1) {
if (interleaved && interleaved <= 2*total_size) {
struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
/* All uploads are interleaved, so upload the arrays together as
* interleaved. First, upload the contents and set up upload[0].
*/
copy_array_to_vbo_array(brw, upload[0], min_index, max_index,
buffer, interleaved);
buffer->offset -= delta * interleaved;
for (i = 0; i < nr_uploads; i++) {
/* Then, just point upload[i] at upload[0]'s buffer. */
upload[i]->offset =
((const unsigned char *)upload[i]->glarray->Ptr - ptr);
upload[i]->buffer = j;
}
j++;
nr_uploads = 0;
}
else if ((total_size < 2048) && can_merge_uploads) {
/* Upload non-interleaved arrays into a single interleaved array */
struct brw_vertex_buffer *buffer;
int count = MAX2(max_index - min_index + 1, 1);
int offset;
char *map;
map = intel_upload_map(&brw->intel, total_size * count, total_size);
for (i = offset = 0; i < nr_uploads; i++) {
const unsigned char *src = upload[i]->glarray->Ptr;
int size = upload[i]->element_size;
int stride = upload[i]->glarray->StrideB;
char *dst;
int n;
offset = ALIGN(offset, get_size(upload[i]->glarray->Type));
dst = map + offset;
src += min_index * stride;
for (n = 0; n < count; n++) {
memcpy(dst, src, size);
src += stride;
dst += total_size;
}
upload[i]->offset = offset;
upload[i]->buffer = j;
offset += size;
}
assert(offset == total_size);
buffer = &brw->vb.buffers[j++];
intel_upload_unmap(&brw->intel, map, offset * count, offset,
&buffer->bo, &buffer->offset);
buffer->stride = offset;
buffer->step_rate = 0;
buffer->offset -= delta * offset;
nr_uploads = 0;
}
}
/* Upload non-interleaved arrays */
for (i = 0; i < nr_uploads; i++) {
struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
if (upload[i]->glarray->InstanceDivisor == 0) {
copy_array_to_vbo_array(brw, upload[i], min_index, max_index,
buffer, upload[i]->element_size);
} else {
/* This is an instanced attribute, since its InstanceDivisor
* is not zero. Therefore, its data will be stepped after the
* instanced draw has been run InstanceDivisor times.
*/
uint32_t instanced_attr_max_index =
(brw->num_instances - 1) / upload[i]->glarray->InstanceDivisor;
copy_array_to_vbo_array(brw, upload[i], 0, instanced_attr_max_index,
buffer, upload[i]->element_size);
}
buffer->offset -= delta * buffer->stride;
buffer->step_rate = upload[i]->glarray->InstanceDivisor;
upload[i]->buffer = j++;
upload[i]->offset = 0;
}
/* can we simply extend the current vb? */
if (j == brw->vb.nr_current_buffers) {
int delta = 0;
for (i = 0; i < j; i++) {
int d;
if (brw->vb.current_buffers[i].handle != brw->vb.buffers[i].bo->handle ||
brw->vb.current_buffers[i].stride != brw->vb.buffers[i].stride ||
brw->vb.current_buffers[i].step_rate != brw->vb.buffers[i].step_rate)
break;
d = brw->vb.buffers[i].offset - brw->vb.current_buffers[i].offset;
if (d < 0)
break;
if (i == 0)
delta = d / brw->vb.current_buffers[i].stride;
if (delta * brw->vb.current_buffers[i].stride != d)
break;
}
if (i == j) {
brw->vb.start_vertex_bias += delta;
while (--j >= 0)
drm_intel_bo_unreference(brw->vb.buffers[j].bo);
j = 0;
}
}
brw->vb.nr_buffers = j;
prepare:
brw_prepare_query_begin(brw);
}
