static void
intel_bufferobj_copy_subdata(GLcontext *ctx,
struct gl_buffer_object *src,
struct gl_buffer_object *dst,
GLintptr read_offset, GLintptr write_offset,
GLsizeiptr size)
{
struct intel_context *intel = intel_context(ctx);
struct intel_buffer_object *intel_src = intel_buffer_object(src);
struct intel_buffer_object *intel_dst = intel_buffer_object(dst);
drm_intel_bo *src_bo, *dst_bo;
if (size == 0)
return;
/* If we're in system memory, just map and memcpy. */
if (intel_src->sys_buffer || intel_dst->sys_buffer) {
/* The same buffer may be used, but note that regions copied may
* not overlap.
*/
if (src == dst) {
char *ptr = intel_bufferobj_map(ctx, GL_COPY_WRITE_BUFFER,
GL_READ_WRITE, dst);
memcpy(ptr + write_offset, ptr + read_offset, size);
intel_bufferobj_unmap(ctx, GL_COPY_WRITE_BUFFER, dst);
} else {
const char *src_ptr;
char *dst_ptr;
src_ptr = intel_bufferobj_map(ctx, GL_COPY_READ_BUFFER,
GL_READ_ONLY, src);
dst_ptr = intel_bufferobj_map(ctx, GL_COPY_WRITE_BUFFER,
GL_WRITE_ONLY, dst);
memcpy(dst_ptr + write_offset, src_ptr + read_offset, size);
intel_bufferobj_unmap(ctx, GL_COPY_READ_BUFFER, src);
intel_bufferobj_unmap(ctx, GL_COPY_WRITE_BUFFER, dst);
}
}
/* Otherwise, we have real BOs, so blit them. */
dst_bo = intel_bufferobj_buffer(intel, intel_dst, INTEL_WRITE_PART);
src_bo = intel_bufferobj_buffer(intel, intel_src, INTEL_READ);
intel_emit_linear_blit(intel,
dst_bo, write_offset,
src_bo, read_offset, size);
/* Since we've emitted some blits to buffers that will (likely) be used
* in rendering operations in other cache domains in this batch, emit a
* flush. Once again, we wish for a domain tracker in libdrm to cover
* usage inside of a batchbuffer.
*/
intel_batchbuffer_emit_mi_flush(intel->batch);
}
/* Attach to a pbo, discarding our data. Effectively zero-copy upload
* the pbo's data.
*/
void
intel_region_attach_pbo(struct intel_context *intel,
struct intel_region *region,
struct intel_buffer_object *pbo)
{
dri_bo *buffer;
if (region->pbo == pbo)
return;
_DBG("%s %p %p\n", __FUNCTION__, region, pbo);
/* If there is already a pbo attached, break the cow tie now.
* Don't call intel_region_release_pbo() as that would
* unnecessarily allocate a new buffer we would have to immediately
* discard.
*/
if (region->pbo) {
region->pbo->region = NULL;
region->pbo = NULL;
}
if (region->buffer) {
dri_bo_unreference(region->buffer);
region->buffer = NULL;
}
/* make sure pbo has a buffer of its own */
buffer = intel_bufferobj_buffer(intel, pbo, INTEL_WRITE_FULL);
region->pbo = pbo;
region->pbo->region = region;
dri_bo_reference(buffer);
region->buffer = buffer;
}
static GLenum
intel_buffer_object_purgeable(struct gl_context * ctx,
struct gl_buffer_object *obj,
GLenum option)
{
struct intel_buffer_object *intel_obj = intel_buffer_object (obj);
if (intel_obj->buffer != NULL)
return intel_buffer_purgeable(intel_obj->buffer);
if (option == GL_RELEASED_APPLE) {
if (intel_obj->sys_buffer != NULL) {
free(intel_obj->sys_buffer);
intel_obj->sys_buffer = NULL;
}
return GL_RELEASED_APPLE;
} else {
/* XXX Create the buffer and madvise(MADV_DONTNEED)? */
struct intel_context *intel = intel_context(ctx);
drm_intel_bo *bo = intel_bufferobj_buffer(intel, intel_obj, INTEL_READ);
return intel_buffer_purgeable(bo);
}
}
void brw_draw_init( struct brw_context *brw )
{
GLcontext *ctx = &brw->intel.ctx;
struct vbo_context *vbo = vbo_context(ctx);
GLuint i;
/* Register our drawing function:
*/
vbo->draw_prims = brw_draw_prims;
brw->vb.upload.size = BRW_UPLOAD_INIT_SIZE;
for (i = 0; i < BRW_NR_UPLOAD_BUFS; i++) {
brw->vb.upload.vbo[i] = ctx->Driver.NewBufferObject(ctx, 1, GL_ARRAY_BUFFER_ARB);
/* NOTE: These are set to no-backing-store.
*/
bmBufferSetInvalidateCB(&brw->intel,
intel_bufferobj_buffer(intel_buffer_object(brw->vb.upload.vbo[i])),
brw_invalidate_vbo_cb,
&brw->intel,
GL_TRUE);
}
ctx->Driver.BufferData( ctx,
GL_ARRAY_BUFFER_ARB,
BRW_UPLOAD_INIT_SIZE,
NULL,
GL_DYNAMIC_DRAW_ARB,
brw->vb.upload.vbo[0] );
}
static bool
intel_set_texture_storage_for_buffer_object(struct gl_context *ctx,
struct gl_texture_object *tex_obj,
struct gl_buffer_object *buffer_obj,
uint32_t buffer_offset,
uint32_t row_stride,
bool read_only)
{
struct brw_context *brw = brw_context(ctx);
struct intel_texture_object *intel_texobj = intel_texture_object(tex_obj);
struct gl_texture_image *image = tex_obj->Image[0][0];
struct intel_texture_image *intel_image = intel_texture_image(image);
struct intel_buffer_object *intel_buffer_obj = intel_buffer_object(buffer_obj);
if (!read_only) {
/* Renderbuffers have the restriction that the buffer offset and
* surface pitch must be a multiple of the element size. If it's
* not, we have to fail and fall back to software.
*/
int cpp = _mesa_get_format_bytes(image->TexFormat);
if (buffer_offset % cpp || row_stride % cpp) {
perf_debug("Bad PBO alignment; fallback to CPU mapping\n");
return false;
}
if (!brw->format_supported_as_render_target[image->TexFormat]) {
perf_debug("Non-renderable PBO format; fallback to CPU mapping\n");
return false;
}
}
assert(intel_texobj->mt == NULL);
drm_intel_bo *bo = intel_bufferobj_buffer(brw, intel_buffer_obj,
buffer_offset,
row_stride * image->Height);
intel_texobj->mt =
intel_miptree_create_for_bo(brw, bo,
image->TexFormat,
buffer_offset,
image->Width, image->Height, image->Depth,
row_stride,
0);
if (!intel_texobj->mt)
return false;
if (!_swrast_init_texture_image(image))
return false;
intel_miptree_reference(&intel_image->mt, intel_texobj->mt);
/* The miptree is in a validated state, so no need to check later. */
intel_texobj->needs_validate = false;
intel_texobj->validated_first_level = 0;
intel_texobj->validated_last_level = 0;
intel_texobj->_Format = intel_texobj->mt->format;
return true;
}
static void
gen8_upload_3dstate_so_buffers(struct brw_context *brw)
{
struct gl_context *ctx = &brw->ctx;
/* BRW_NEW_TRANSFORM_FEEDBACK */
struct gl_transform_feedback_object *xfb_obj =
ctx->TransformFeedback.CurrentObject;
struct brw_transform_feedback_object *brw_obj =
(struct brw_transform_feedback_object *) xfb_obj;
/* Set up the up to 4 output buffers. These are the ranges defined in the
* gl_transform_feedback_object.
*/
for (int i = 0; i < 4; i++) {
struct intel_buffer_object *bufferobj =
intel_buffer_object(xfb_obj->Buffers[i]);
if (!bufferobj) {
BEGIN_BATCH(8);
OUT_BATCH(_3DSTATE_SO_BUFFER << 16 | (8 - 2));
OUT_BATCH((i << SO_BUFFER_INDEX_SHIFT));
OUT_BATCH(0);
OUT_BATCH(0);
OUT_BATCH(0);
OUT_BATCH(0);
OUT_BATCH(0);
OUT_BATCH(0);
ADVANCE_BATCH();
continue;
}
uint32_t start = xfb_obj->Offset[i];
assert(start % 4 == 0);
uint32_t end = ALIGN(start + xfb_obj->Size[i], 4);
drm_intel_bo *bo =
intel_bufferobj_buffer(brw, bufferobj, start, end - start);
assert(end <= bo->size);
perf_debug("Missing MOCS setup for 3DSTATE_SO_BUFFER.");
BEGIN_BATCH(8);
OUT_BATCH(_3DSTATE_SO_BUFFER << 16 | (8 - 2));
OUT_BATCH(GEN8_SO_BUFFER_ENABLE | (i << SO_BUFFER_INDEX_SHIFT) |
GEN8_SO_BUFFER_OFFSET_WRITE_ENABLE |
GEN8_SO_BUFFER_OFFSET_ADDRESS_ENABLE |
(BDW_MOCS_WB << 22));
OUT_RELOC64(bo, I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, start);
OUT_BATCH(xfb_obj->Size[i] / 4 - 1);
OUT_RELOC64(brw_obj->offset_bo,
I915_GEM_DOMAIN_INSTRUCTION, I915_GEM_DOMAIN_INSTRUCTION,
i * sizeof(uint32_t));
if (brw_obj->zero_offsets)
OUT_BATCH(0); /* Zero out the offset and write that to offset_bo */
else
OUT_BATCH(0xFFFFFFFF); /* Use offset_bo as the "Stream Offset." */
ADVANCE_BATCH();
}
brw_obj->zero_offsets = false;
}
/* XXX: Do this for TexSubImage also:
*/
static GLboolean
try_pbo_upload(struct intel_context *intel,
struct intel_texture_image *intelImage,
const struct gl_pixelstore_attrib *unpack,
GLint internalFormat,
GLint width, GLint height,
GLenum format, GLenum type, const void *pixels)
{
struct intel_buffer_object *pbo = intel_buffer_object(unpack->BufferObj);
GLuint src_offset, src_stride;
GLuint dst_offset, dst_stride;
if (!pbo ||
intel->ctx._ImageTransferState ||
unpack->SkipPixels || unpack->SkipRows) {
_mesa_printf("%s: failure 1\n", __FUNCTION__);
return GL_FALSE;
}
src_offset = (GLuint) pixels;
if (unpack->RowLength > 0)
src_stride = unpack->RowLength;
else
src_stride = width;
dst_offset = intel_miptree_image_offset(intelImage->mt,
intelImage->face,
intelImage->level);
dst_stride = intelImage->mt->pitch;
intelFlush(&intel->ctx);
LOCK_HARDWARE(intel);
{
struct _DriBufferObject *src_buffer =
intel_bufferobj_buffer(intel, pbo, INTEL_READ);
struct _DriBufferObject *dst_buffer =
intel_region_buffer(intel->intelScreen, intelImage->mt->region,
INTEL_WRITE_FULL);
intelEmitCopyBlit(intel,
intelImage->mt->cpp,
src_stride, src_buffer, src_offset,
dst_stride, dst_buffer, dst_offset,
0, 0, 0, 0, width, height,
GL_COPY);
intel_batchbuffer_flush(intel->batch);
}
UNLOCK_HARDWARE(intel);
return GL_TRUE;
}
static void
upload_3dstate_so_buffers(struct brw_context *brw)
{
struct gl_context *ctx = &brw->ctx;
/* BRW_NEW_VERTEX_PROGRAM */
const struct gl_shader_program *vs_prog =
ctx->Shader.CurrentVertexProgram;
const struct gl_transform_feedback_info *linked_xfb_info =
&vs_prog->LinkedTransformFeedback;
/* BRW_NEW_TRANSFORM_FEEDBACK */
struct gl_transform_feedback_object *xfb_obj =
ctx->TransformFeedback.CurrentObject;
int i;
/* Set up the up to 4 output buffers. These are the ranges defined in the
* gl_transform_feedback_object.
*/
for (i = 0; i < 4; i++) {
struct intel_buffer_object *bufferobj =
intel_buffer_object(xfb_obj->Buffers[i]);
drm_intel_bo *bo;
uint32_t start, end;
uint32_t stride;
if (!xfb_obj->Buffers[i]) {
/* The pitch of 0 in this command indicates that the buffer is
* unbound and won't be written to.
*/
BEGIN_BATCH(4);
OUT_BATCH(_3DSTATE_SO_BUFFER << 16 | (4 - 2));
OUT_BATCH((i << SO_BUFFER_INDEX_SHIFT));
OUT_BATCH(0);
OUT_BATCH(0);
ADVANCE_BATCH();
continue;
}
bo = intel_bufferobj_buffer(brw, bufferobj, INTEL_WRITE_PART);
stride = linked_xfb_info->BufferStride[i] * 4;
start = xfb_obj->Offset[i];
assert(start % 4 == 0);
end = ALIGN(start + xfb_obj->Size[i], 4);
assert(end <= bo->size);
BEGIN_BATCH(4);
OUT_BATCH(_3DSTATE_SO_BUFFER << 16 | (4 - 2));
OUT_BATCH((i << SO_BUFFER_INDEX_SHIFT) | stride);
OUT_RELOC(bo, I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, start);
OUT_RELOC(bo, I915_GEM_DOMAIN_RENDER, I915_GEM_DOMAIN_RENDER, end);
ADVANCE_BATCH();
}
}
int brw_prepare_indices( struct brw_context *brw,
const struct _mesa_index_buffer *index_buffer,
dri_bo **bo_return,
GLuint *offset_return)
{
GLcontext *ctx = &brw->intel.ctx;
struct intel_context *intel = &brw->intel;
GLuint ib_size = get_size(index_buffer->type) * index_buffer->count;
dri_bo *bo;
struct gl_buffer_object *bufferobj = index_buffer->obj;
GLuint offset = (GLuint)index_buffer->ptr;
int ret;
/* Turn into a proper VBO:
*/
if (!bufferobj->Name) {
/* Get new bufferobj, offset:
*/
get_space(brw, ib_size, &bo, &offset);
/* Straight upload
*/
dri_bo_subdata(bo, offset, ib_size, index_buffer->ptr);
} else {
/* If the index buffer isn't aligned to its element size, we have to
* rebase it into a temporary.
*/
if ((get_size(index_buffer->type) - 1) & offset) {
GLubyte *map = ctx->Driver.MapBuffer(ctx,
GL_ELEMENT_ARRAY_BUFFER_ARB,
GL_DYNAMIC_DRAW_ARB,
bufferobj);
map += offset;
get_space(brw, ib_size, &bo, &offset);
dri_bo_subdata(bo, offset, ib_size, map);
ctx->Driver.UnmapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER_ARB, bufferobj);
} else {
bo = intel_bufferobj_buffer(intel, intel_buffer_object(bufferobj),
INTEL_READ);
dri_bo_reference(bo);
}
}
*bo_return = bo;
*offset_return = offset;
ret = dri_bufmgr_check_aperture_space(bo);
return ret;
}
static void brw_upload_indices(struct brw_context *brw)
{
struct gl_context *ctx = &brw->ctx;
const struct _mesa_index_buffer *index_buffer = brw->ib.ib;
GLuint ib_size;
drm_intel_bo *old_bo = brw->ib.bo;
struct gl_buffer_object *bufferobj;
GLuint offset;
GLuint ib_type_size;
if (index_buffer == NULL)
return;
ib_type_size = _mesa_sizeof_type(index_buffer->type);
ib_size = ib_type_size * index_buffer->count;
bufferobj = index_buffer->obj;
/* Turn into a proper VBO:
*/
if (!_mesa_is_bufferobj(bufferobj)) {
/* Get new bufferobj, offset:
*/
intel_upload_data(brw, index_buffer->ptr, ib_size, ib_type_size,
&brw->ib.bo, &offset);
} else {
offset = (GLuint) (unsigned long) index_buffer->ptr;
/* If the index buffer isn't aligned to its element size, we have to
* rebase it into a temporary.
*/
if ((ib_type_size - 1) & offset) {
perf_debug("copying index buffer to a temporary to work around "
"misaligned offset %d\n", offset);
GLubyte *map = ctx->Driver.MapBufferRange(ctx,
offset,
ib_size,
GL_MAP_READ_BIT,
bufferobj,
MAP_INTERNAL);
intel_upload_data(brw, map, ib_size, ib_type_size,
&brw->ib.bo, &offset);
ctx->Driver.UnmapBuffer(ctx, bufferobj, MAP_INTERNAL);
} else {
drm_intel_bo *bo =
intel_bufferobj_buffer(brw, intel_buffer_object(bufferobj),
offset, ib_size);
if (bo != brw->ib.bo) {
drm_intel_bo_unreference(brw->ib.bo);
brw->ib.bo = bo;
drm_intel_bo_reference(bo);
}
}
}
/* Use 3DPRIMITIVE's start_vertex_offset to avoid re-uploading
* the index buffer state when we're just moving the start index
* of our drawing.
*/
brw->ib.start_vertex_offset = offset / ib_type_size;
if (brw->ib.bo != old_bo)
brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER;
if (index_buffer->type != brw->ib.type) {
brw->ib.type = index_buffer->type;
brw->state.dirty.brw |= BRW_NEW_INDEX_BUFFER;
}
}
void
brw_prepare_vertices(struct brw_context *brw)
{
struct gl_context *ctx = &brw->ctx;
/* CACHE_NEW_VS_PROG */
GLbitfield64 vs_inputs = brw->vs.prog_data->inputs_read;
const unsigned char *ptr = NULL;
GLuint interleaved = 0;
unsigned int min_index = brw->vb.min_index + brw->basevertex;
unsigned int max_index = brw->vb.max_index + brw->basevertex;
int delta, i, j;
struct brw_vertex_element *upload[VERT_ATTRIB_MAX];
GLuint nr_uploads = 0;
/* _NEW_POLYGON
*
* On gen6+, edge flags don't end up in the VUE (either in or out of the
* VS). Instead, they're uploaded as the last vertex element, and the data
* is passed sideband through the fixed function units. So, we need to
* prepare the vertex buffer for it, but it's not present in inputs_read.
*/
if (brw->gen >= 6 && (ctx->Polygon.FrontMode != GL_FILL ||
ctx->Polygon.BackMode != GL_FILL)) {
vs_inputs |= VERT_BIT_EDGEFLAG;
}
if (0)
fprintf(stderr, "%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);
brw->vb.enabled[brw->vb.nr_enabled++] = input;
}
if (brw->vb.nr_enabled == 0)
return;
if (brw->vb.nr_buffers)
return;
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;
if (_mesa_is_bufferobj(glarray->BufferObj)) {
struct intel_buffer_object *intel_buffer =
intel_buffer_object(glarray->BufferObj);
int k;
/* If we have a VB set to be uploaded for this buffer object
* already, reuse that VB state so that we emit fewer
* relocations.
*/
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->offset = (uintptr_t)glarray->Ptr;
buffer->stride = glarray->StrideB;
buffer->step_rate = glarray->InstanceDivisor;
uint32_t offset, size;
if (glarray->InstanceDivisor) {
offset = buffer->offset;
size = (buffer->stride * ((brw->num_instances /
glarray->InstanceDivisor) - 1) +
glarray->_ElementSize);
} else {
if (min_index == -1) {
offset = 0;
size = intel_buffer->Base.Size;
} else {
offset = buffer->offset + min_index * buffer->stride;
size = (buffer->stride * (max_index - min_index) +
glarray->_ElementSize);
}
}
buffer->bo = intel_bufferobj_buffer(brw, intel_buffer,
offset, size);
drm_intel_bo_reference(buffer->bo);
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) {
interleaved = glarray->StrideB;
ptr = glarray->Ptr;
}
else if (interleaved != glarray->StrideB ||
glarray->Ptr < ptr ||
(uintptr_t)(glarray->Ptr - ptr) + glarray->_ElementSize > interleaved)
{
/* If our stride is different from the first attribute's stride,
* or if the first attribute's stride didn't cover our element,
* disable the interleaved upload optimization. The second case
* can most commonly occur in cases where there is a single vertex
* and, for example, the data is stored on the application's
* stack.
*
* NOTE: This will also disable the optimization in cases where
* the data is in a different order than the array indices.
* Something like:
*
* float data[...];
* glVertexAttribPointer(0, 4, GL_FLOAT, 32, &data[4]);
* glVertexAttribPointer(1, 4, GL_FLOAT, 32, &data[0]);
*/
interleaved = 0;
}
upload[nr_uploads++] = input;
}
}
/* 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;
}
/* Handle any arrays to be uploaded. */
if (nr_uploads > 1) {
if (interleaved) {
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;
}
}
/* 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]->glarray->_ElementSize);
} 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]->glarray->_ElementSize);
}
buffer->offset -= delta * buffer->stride;
buffer->step_rate = upload[i]->glarray->InstanceDivisor;
upload[i]->buffer = j++;
upload[i]->offset = 0;
}
brw->vb.nr_buffers = j;
}
static bool
do_blit_drawpixels(struct gl_context * ctx,
GLint x, GLint y, GLsizei width, GLsizei height,
GLenum format, GLenum type,
const struct gl_pixelstore_attrib *unpack,
const GLvoid * pixels)
{
struct brw_context *brw = brw_context(ctx);
struct intel_buffer_object *src = intel_buffer_object(unpack->BufferObj);
GLuint src_offset;
drm_intel_bo *src_buffer;
DBG("%s\n", __FUNCTION__);
if (!intel_check_blit_fragment_ops(ctx, false))
return false;
if (ctx->DrawBuffer->_NumColorDrawBuffers != 1) {
DBG("%s: fallback due to MRT\n", __FUNCTION__);
return false;
}
struct gl_renderbuffer *rb = ctx->DrawBuffer->_ColorDrawBuffers[0];
struct intel_renderbuffer *irb = intel_renderbuffer(rb);
if (!_mesa_format_matches_format_and_type(irb->mt->format, format, type,
false)) {
DBG("%s: bad format for blit\n", __FUNCTION__);
return false;
}
if (unpack->SwapBytes || unpack->LsbFirst ||
unpack->SkipPixels || unpack->SkipRows) {
DBG("%s: bad packing params\n", __FUNCTION__);
return false;
}
int src_stride = _mesa_image_row_stride(unpack, width, format, type);
bool src_flip = false;
/* Mesa flips the src_stride for unpack->Invert, but we want our mt to have
* a normal src_stride.
*/
if (unpack->Invert) {
src_stride = -src_stride;
src_flip = true;
}
src_offset = (GLintptr)pixels;
src_offset += _mesa_image_offset(2, unpack, width, height,
format, type, 0, 0, 0);
intel_prepare_render(brw);
src_buffer = intel_bufferobj_buffer(brw, src,
src_offset, width * height *
irb->mt->cpp);
struct intel_mipmap_tree *pbo_mt =
intel_miptree_create_for_bo(brw,
src_buffer,
irb->mt->format,
src_offset,
width, height,
src_stride, I915_TILING_NONE);
if (!pbo_mt)
return false;
if (!intel_miptree_blit(brw,
pbo_mt, 0, 0,
0, 0, src_flip,
irb->mt, irb->mt_level, irb->mt_layer,
x, y, _mesa_is_winsys_fbo(ctx->DrawBuffer),
width, height, GL_COPY)) {
DBG("%s: blit failed\n", __FUNCTION__);
intel_miptree_release(&pbo_mt);
return false;
}
intel_miptree_release(&pbo_mt);
if (ctx->Query.CurrentOcclusionObject)
ctx->Query.CurrentOcclusionObject->Result += width * height;
intel_check_front_buffer_rendering(brw);
DBG("%s: success\n", __FUNCTION__);
return true;
}
static GLboolean
do_blit_readpixels(GLcontext * ctx,
GLint x, GLint y, GLsizei width, GLsizei height,
GLenum format, GLenum type,
const struct gl_pixelstore_attrib *pack, GLvoid * pixels)
{
struct intel_context *intel = intel_context(ctx);
struct intel_region *src = intel_readbuf_region(intel);
struct intel_buffer_object *dst = intel_buffer_object(pack->BufferObj);
GLuint dst_offset;
GLuint rowLength;
drm_intel_bo *dst_buffer;
GLboolean all;
GLint dst_x, dst_y;
if (INTEL_DEBUG & DEBUG_PIXEL)
printf("%s\n", __FUNCTION__);
if (!src)
return GL_FALSE;
if (!_mesa_is_bufferobj(pack->BufferObj)) {
/* PBO only for now:
*/
if (INTEL_DEBUG & DEBUG_PIXEL)
printf("%s - not PBO\n", __FUNCTION__);
return GL_FALSE;
}
if (ctx->_ImageTransferState ||
!intel_check_blit_format(src, format, type)) {
if (INTEL_DEBUG & DEBUG_PIXEL)
printf("%s - bad format for blit\n", __FUNCTION__);
return GL_FALSE;
}
if (pack->Alignment != 1 || pack->SwapBytes || pack->LsbFirst) {
if (INTEL_DEBUG & DEBUG_PIXEL)
printf("%s: bad packing params\n", __FUNCTION__);
return GL_FALSE;
}
if (pack->RowLength > 0)
rowLength = pack->RowLength;
else
rowLength = width;
if (pack->Invert) {
if (INTEL_DEBUG & DEBUG_PIXEL)
printf("%s: MESA_PACK_INVERT not done yet\n", __FUNCTION__);
return GL_FALSE;
}
else {
if (ctx->ReadBuffer->Name == 0)
rowLength = -rowLength;
}
dst_offset = (GLintptr) _mesa_image_address(2, pack, pixels, width, height,
format, type, 0, 0, 0);
if (!_mesa_clip_copytexsubimage(ctx,
&dst_x, &dst_y,
&x, &y,
&width, &height)) {
return GL_TRUE;
}
intel_prepare_render(intel);
all = (width * height * src->cpp == dst->Base.Size &&
x == 0 && dst_offset == 0);
dst_x = 0;
dst_y = 0;
dst_buffer = intel_bufferobj_buffer(intel, dst,
all ? INTEL_WRITE_FULL :
INTEL_WRITE_PART);
if (ctx->ReadBuffer->Name == 0)
y = ctx->ReadBuffer->Height - (y + height);
if (!intelEmitCopyBlit(intel,
src->cpp,
src->pitch, src->buffer, 0, src->tiling,
rowLength, dst_buffer, dst_offset, GL_FALSE,
x, y,
dst_x, dst_y,
width, height,
GL_COPY)) {
return GL_FALSE;
}
if (INTEL_DEBUG & DEBUG_PIXEL)
printf("%s - DONE\n", __FUNCTION__);
return GL_TRUE;
}
int brw_prepare_vertices( struct brw_context *brw,
GLuint min_index,
GLuint max_index )
{
GLcontext *ctx = &brw->intel.ctx;
struct intel_context *intel = intel_context(ctx);
GLuint tmp = brw->vs.prog_data->inputs_read;
GLuint i;
const unsigned char *ptr = NULL;
GLuint interleave = 0;
int ret = 0;
struct brw_vertex_element *enabled[VERT_ATTRIB_MAX];
GLuint nr_enabled = 0;
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)
_mesa_printf("%s %d..%d\n", __FUNCTION__, min_index, max_index);
/* Accumulate the list of enabled arrays. */
while (tmp) {
GLuint i = _mesa_ffsll(tmp)-1;
struct brw_vertex_element *input = &brw->vb.inputs[i];
tmp &= ~(1<<i);
enabled[nr_enabled++] = input;
}
/* XXX: In the rare cases where this happens we fallback all
* the way to software rasterization, although a tnl fallback
* would be sufficient. I don't know of *any* real world
* cases with > 17 vertex attributes enabled, so it probably
* isn't an issue at this point.
*/
if (nr_enabled >= BRW_VEP_MAX)
return -1;
for (i = 0; i < nr_enabled; i++) {
struct brw_vertex_element *input = enabled[i];
input->element_size = get_size(input->glarray->Type) * input->glarray->Size;
input->count = input->glarray->StrideB ? max_index + 1 - min_index : 1;
if (input->glarray->BufferObj->Name != 0) {
struct intel_buffer_object *intel_buffer =
intel_buffer_object(input->glarray->BufferObj);
/* Named buffer object: Just reference its contents directly. */
input->bo = intel_bufferobj_buffer(intel, intel_buffer,
INTEL_READ);
dri_bo_reference(input->bo);
input->offset = (unsigned long)input->glarray->Ptr;
input->stride = input->glarray->StrideB;
ret |= dri_bufmgr_check_aperture_space(input->bo);
} 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 (i == 0) {
/* Position array not properly enabled:
*/
if (input->glarray->StrideB == 0)
return -1;
interleave = input->glarray->StrideB;
ptr = input->glarray->Ptr;
}
else if (interleave != input->glarray->StrideB ||
(const unsigned char *)input->glarray->Ptr - ptr < 0 ||
(const unsigned char *)input->glarray->Ptr - ptr > interleave)
{
interleave = 0;
}
upload[nr_uploads++] = input;
/* We rebase drawing to start at element zero only when
* varyings are not in vbos, which means we can end up
* uploading non-varying arrays (stride != 0) when min_index
* is zero. This doesn't matter as the amount to upload is
* the same for these arrays whether the draw call is rebased
* or not - we just have to upload the one element.
*/
assert(min_index == 0 || input->glarray->StrideB == 0);
}
}
/* Handle any arrays to be uploaded. */
if (nr_uploads > 1 && interleave && interleave <= 256) {
/* 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], interleave);
ret |= dri_bufmgr_check_aperture_space(upload[0]->bo);
for (i = 1; i < nr_uploads; i++) {
/* Then, just point upload[i] at upload[0]'s buffer. */
upload[i]->stride = interleave;
upload[i]->offset = upload[0]->offset +
((const unsigned char *)upload[i]->glarray->Ptr - ptr);
upload[i]->bo = upload[0]->bo;
dri_bo_reference(upload[i]->bo);
}
}
else {
/* Upload non-interleaved arrays */
for (i = 0; i < nr_uploads; i++) {
copy_array_to_vbo_array(brw, upload[i], upload[i]->element_size);
if (upload[i]->bo) {
ret |= dri_bufmgr_check_aperture_space(upload[i]->bo);
}
}
}
if (ret)
return 1;
return 0;
}
static struct buffer *array_buffer( const struct gl_client_array *array )
{
return intel_bufferobj_buffer(intel_buffer_object(array->BufferObj));
}
void brw_upload_indices( struct brw_context *brw,
const struct _mesa_index_buffer *index_buffer )
{
GLcontext *ctx = &brw->intel.ctx;
struct intel_context *intel = &brw->intel;
GLuint ib_size = get_size(index_buffer->type) * index_buffer->count;
struct gl_buffer_object *bufferobj = index_buffer->obj;
GLuint offset = (GLuint)index_buffer->ptr;
/* Turn into a proper VBO:
*/
if (!bufferobj->Name) {
/* Get new bufferobj, offset:
*/
get_space(brw, ib_size, &bufferobj, &offset);
/* Straight upload
*/
ctx->Driver.BufferSubData( ctx,
GL_ELEMENT_ARRAY_BUFFER_ARB,
offset,
ib_size,
index_buffer->ptr,
bufferobj);
} else {
/* If the index buffer isn't aligned to its element size, we have to
* rebase it into a temporary.
*/
if ((get_size(index_buffer->type) - 1) & offset) {
struct gl_buffer_object *vbo;
GLuint voffset;
GLubyte *map = ctx->Driver.MapBuffer(ctx,
GL_ELEMENT_ARRAY_BUFFER_ARB,
GL_DYNAMIC_DRAW_ARB,
bufferobj);
map += offset;
get_space(brw, ib_size, &vbo, &voffset);
ctx->Driver.BufferSubData(ctx,
GL_ELEMENT_ARRAY_BUFFER_ARB,
voffset,
ib_size,
map,
vbo);
ctx->Driver.UnmapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER_ARB, bufferobj);
bufferobj = vbo;
offset = voffset;
}
}
/* Emit the indexbuffer packet:
*/
{
struct brw_indexbuffer ib;
struct buffer *buffer = intel_bufferobj_buffer(intel_buffer_object(bufferobj));
memset(&ib, 0, sizeof(ib));
ib.header.bits.opcode = CMD_INDEX_BUFFER;
ib.header.bits.length = sizeof(ib)/4 - 2;
ib.header.bits.index_format = get_index_type(index_buffer->type);
ib.header.bits.cut_index_enable = 0;
BEGIN_BATCH(4, 0);
OUT_BATCH( ib.header.dword );
OUT_BATCH( bmBufferOffset(intel, buffer) + offset );
OUT_BATCH( bmBufferOffset(intel, buffer) + offset + ib_size );
OUT_BATCH( 0 );
ADVANCE_BATCH();
}
}
static bool
do_blit_readpixels(struct gl_context * ctx,
GLint x, GLint y, GLsizei width, GLsizei height,
GLenum format, GLenum type,
const struct gl_pixelstore_attrib *pack, GLvoid * pixels)
{
struct brw_context *brw = brw_context(ctx);
struct intel_buffer_object *dst = intel_buffer_object(pack->BufferObj);
GLuint dst_offset;
drm_intel_bo *dst_buffer;
GLint dst_x, dst_y;
GLuint dirty;
DBG("%s\n", __FUNCTION__);
assert(_mesa_is_bufferobj(pack->BufferObj));
struct gl_renderbuffer *rb = ctx->ReadBuffer->_ColorReadBuffer;
struct intel_renderbuffer *irb = intel_renderbuffer(rb);
if (ctx->_ImageTransferState ||
!_mesa_format_matches_format_and_type(irb->mt->format, format, type,
false)) {
DBG("%s - bad format for blit\n", __FUNCTION__);
return false;
}
if (pack->SwapBytes || pack->LsbFirst) {
DBG("%s: bad packing params\n", __FUNCTION__);
return false;
}
int dst_stride = _mesa_image_row_stride(pack, width, format, type);
bool dst_flip = false;
/* Mesa flips the dst_stride for pack->Invert, but we want our mt to have a
* normal dst_stride.
*/
struct gl_pixelstore_attrib uninverted_pack = *pack;
if (pack->Invert) {
dst_stride = -dst_stride;
dst_flip = true;
uninverted_pack.Invert = false;
}
dst_offset = (GLintptr)pixels;
dst_offset += _mesa_image_offset(2, &uninverted_pack, width, height,
format, type, 0, 0, 0);
if (!_mesa_clip_copytexsubimage(ctx,
&dst_x, &dst_y,
&x, &y,
&width, &height)) {
return true;
}
dirty = brw->front_buffer_dirty;
intel_prepare_render(brw);
brw->front_buffer_dirty = dirty;
dst_buffer = intel_bufferobj_buffer(brw, dst,
dst_offset, height * dst_stride);
struct intel_mipmap_tree *pbo_mt =
intel_miptree_create_for_bo(brw,
dst_buffer,
irb->mt->format,
dst_offset,
width, height,
dst_stride, I915_TILING_NONE);
if (!intel_miptree_blit(brw,
irb->mt, irb->mt_level, irb->mt_layer,
x, y, _mesa_is_winsys_fbo(ctx->ReadBuffer),
pbo_mt, 0, 0,
0, 0, dst_flip,
width, height, GL_COPY)) {
return false;
}
intel_miptree_release(&pbo_mt);
DBG("%s - DONE\n", __FUNCTION__);
return true;
}
static void
brw_emit_prim(struct brw_context *brw,
const struct _mesa_prim *prim,
uint32_t hw_prim)
{
int verts_per_instance;
int vertex_access_type;
int indirect_flag;
DBG("PRIM: %s %d %d\n", _mesa_enum_to_string(prim->mode),
prim->start, prim->count);
int start_vertex_location = prim->start;
int base_vertex_location = prim->basevertex;
if (prim->indexed) {
vertex_access_type = brw->gen >= 7 ?
GEN7_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM :
GEN4_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM;
start_vertex_location += brw->ib.start_vertex_offset;
base_vertex_location += brw->vb.start_vertex_bias;
} else {
vertex_access_type = brw->gen >= 7 ?
GEN7_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL :
GEN4_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL;
start_vertex_location += brw->vb.start_vertex_bias;
}
/* We only need to trim the primitive count on pre-Gen6. */
if (brw->gen < 6)
verts_per_instance = trim(prim->mode, prim->count);
else
verts_per_instance = prim->count;
/* If nothing to emit, just return. */
if (verts_per_instance == 0 && !prim->is_indirect)
return;
/* If we're set to always flush, do it before and after the primitive emit.
* We want to catch both missed flushes that hurt instruction/state cache
* and missed flushes of the render cache as it heads to other parts of
* the besides the draw code.
*/
if (brw->always_flush_cache)
brw_emit_mi_flush(brw);
/* If indirect, emit a bunch of loads from the indirect BO. */
if (prim->is_indirect) {
struct gl_buffer_object *indirect_buffer = brw->ctx.DrawIndirectBuffer;
drm_intel_bo *bo = intel_bufferobj_buffer(brw,
intel_buffer_object(indirect_buffer),
prim->indirect_offset, 5 * sizeof(GLuint));
indirect_flag = GEN7_3DPRIM_INDIRECT_PARAMETER_ENABLE;
brw_load_register_mem(brw, GEN7_3DPRIM_VERTEX_COUNT, bo,
I915_GEM_DOMAIN_VERTEX, 0,
prim->indirect_offset + 0);
brw_load_register_mem(brw, GEN7_3DPRIM_INSTANCE_COUNT, bo,
I915_GEM_DOMAIN_VERTEX, 0,
prim->indirect_offset + 4);
brw_load_register_mem(brw, GEN7_3DPRIM_START_VERTEX, bo,
I915_GEM_DOMAIN_VERTEX, 0,
prim->indirect_offset + 8);
if (prim->indexed) {
brw_load_register_mem(brw, GEN7_3DPRIM_BASE_VERTEX, bo,
I915_GEM_DOMAIN_VERTEX, 0,
prim->indirect_offset + 12);
brw_load_register_mem(brw, GEN7_3DPRIM_START_INSTANCE, bo,
I915_GEM_DOMAIN_VERTEX, 0,
prim->indirect_offset + 16);
} else {
brw_load_register_mem(brw, GEN7_3DPRIM_START_INSTANCE, bo,
I915_GEM_DOMAIN_VERTEX, 0,
prim->indirect_offset + 12);
BEGIN_BATCH(3);
OUT_BATCH(MI_LOAD_REGISTER_IMM | (3 - 2));
OUT_BATCH(GEN7_3DPRIM_BASE_VERTEX);
OUT_BATCH(0);
ADVANCE_BATCH();
}
} else {
indirect_flag = 0;
}
BEGIN_BATCH(brw->gen >= 7 ? 7 : 6);
if (brw->gen >= 7) {
const int predicate_enable =
(brw->predicate.state == BRW_PREDICATE_STATE_USE_BIT)
? GEN7_3DPRIM_PREDICATE_ENABLE : 0;
OUT_BATCH(CMD_3D_PRIM << 16 | (7 - 2) | indirect_flag | predicate_enable);
OUT_BATCH(hw_prim | vertex_access_type);
} else {
OUT_BATCH(CMD_3D_PRIM << 16 | (6 - 2) |
hw_prim << GEN4_3DPRIM_TOPOLOGY_TYPE_SHIFT |
vertex_access_type);
}
OUT_BATCH(verts_per_instance);
OUT_BATCH(start_vertex_location);
OUT_BATCH(prim->num_instances);
OUT_BATCH(prim->base_instance);
OUT_BATCH(base_vertex_location);
ADVANCE_BATCH();
if (brw->always_flush_cache)
brw_emit_mi_flush(brw);
}
/* Pros:
* - no waiting for idle before updating framebuffer.
*
* Cons:
* - if upload is by memcpy, this may actually be slower than fallback path.
* - uploads the whole image even if destination is clipped
*
* Need to benchmark.
*
* Given the questions about performance, implement for pbo's only.
* This path is definitely a win if the pbo is already in agp. If it
* turns out otherwise, we can add the code necessary to upload client
* data to agp space before performing the blit. (Though it may turn
* out to be better/simpler just to use the texture engine).
*/
static GLboolean
do_blit_drawpixels(GLcontext * ctx,
GLint x, GLint y,
GLsizei width, GLsizei height,
GLenum format, GLenum type,
const struct gl_pixelstore_attrib *unpack,
const GLvoid * pixels)
{
struct intel_context *intel = intel_context(ctx);
struct intel_region *dest = intel_drawbuf_region(intel);
struct intel_buffer_object *src = intel_buffer_object(unpack->BufferObj);
GLuint src_offset;
GLuint rowLength;
struct _DriFenceObject *fence = NULL;
if (INTEL_DEBUG & DEBUG_PIXEL)
_mesa_printf("%s\n", __FUNCTION__);
if (!dest) {
if (INTEL_DEBUG & DEBUG_PIXEL)
_mesa_printf("%s - no dest\n", __FUNCTION__);
return GL_FALSE;
}
if (src) {
/* This validation should be done by core mesa:
*/
if (!_mesa_validate_pbo_access(2, unpack, width, height, 1,
format, type, pixels)) {
_mesa_error(ctx, GL_INVALID_OPERATION, "glDrawPixels");
return GL_TRUE;
}
}
else {
/* PBO only for now:
*/
if (INTEL_DEBUG & DEBUG_PIXEL)
_mesa_printf("%s - not PBO\n", __FUNCTION__);
return GL_FALSE;
}
if (!intel_check_blit_format(dest, format, type)) {
if (INTEL_DEBUG & DEBUG_PIXEL)
_mesa_printf("%s - bad format for blit\n", __FUNCTION__);
return GL_FALSE;
}
if (!intel_check_blit_fragment_ops(ctx)) {
if (INTEL_DEBUG & DEBUG_PIXEL)
_mesa_printf("%s - bad GL fragment state for blitter\n",
__FUNCTION__);
return GL_FALSE;
}
if (ctx->Pixel.ZoomX != 1.0F) {
if (INTEL_DEBUG & DEBUG_PIXEL)
_mesa_printf("%s - bad PixelZoomX for blit\n", __FUNCTION__);
return GL_FALSE;
}
if (unpack->RowLength > 0)
rowLength = unpack->RowLength;
else
rowLength = width;
if (ctx->Pixel.ZoomY == -1.0F) {
if (INTEL_DEBUG & DEBUG_PIXEL)
_mesa_printf("%s - bad PixelZoomY for blit\n", __FUNCTION__);
return GL_FALSE; /* later */
y -= height;
}
else if (ctx->Pixel.ZoomY == 1.0F) {
rowLength = -rowLength;
}
else {
if (INTEL_DEBUG & DEBUG_PIXEL)
_mesa_printf("%s - bad PixelZoomY for blit\n", __FUNCTION__);
return GL_FALSE;
}
src_offset = (GLuint) _mesa_image_address(2, unpack, pixels, width, height,
format, type, 0, 0, 0);
intelFlush(&intel->ctx);
LOCK_HARDWARE(intel);
if (intel->driDrawable->numClipRects) {
__DRIdrawablePrivate *dPriv = intel->driDrawable;
int nbox = dPriv->numClipRects;
drm_clip_rect_t *box = dPriv->pClipRects;
drm_clip_rect_t rect;
drm_clip_rect_t dest_rect;
struct _DriBufferObject *src_buffer =
intel_bufferobj_buffer(intel, src, INTEL_READ);
int i;
dest_rect.x1 = dPriv->x + x;
dest_rect.y1 = dPriv->y + dPriv->h - (y + height);
dest_rect.x2 = dest_rect.x1 + width;
dest_rect.y2 = dest_rect.y1 + height;
for (i = 0; i < nbox; i++) {
if (!intel_intersect_cliprects(&rect, &dest_rect, &box[i]))
continue;
intelEmitCopyBlit(intel,
dest->cpp,
rowLength,
src_buffer, src_offset,
dest->pitch,
dest->buffer, 0,
rect.x1 - dest_rect.x1,
rect.y2 - dest_rect.y2,
rect.x1,
rect.y1, rect.x2 - rect.x1, rect.y2 - rect.y1,
ctx->Color.ColorLogicOpEnabled ?
ctx->Color.LogicOp : GL_COPY);
}
fence = intel_batchbuffer_flush(intel->batch);
driFenceReference(fence);
}
UNLOCK_HARDWARE(intel);
if (fence) {
driFenceFinish(fence, DRM_FENCE_TYPE_EXE | DRM_I915_FENCE_TYPE_RW, GL_FALSE);
driFenceUnReference(fence);
}
if (INTEL_DEBUG & DEBUG_PIXEL)
_mesa_printf("%s - DONE\n", __FUNCTION__);
return GL_TRUE;
}
void
brw_prepare_vertices(struct brw_context *brw)
{
struct gl_context *ctx = &brw->ctx;
/* BRW_NEW_VS_PROG_DATA */
const struct brw_vs_prog_data *vs_prog_data =
brw_vs_prog_data(brw->vs.base.prog_data);
GLbitfield64 vs_inputs = vs_prog_data->inputs_read;
const unsigned char *ptr = NULL;
GLuint interleaved = 0;
unsigned int min_index = brw->vb.min_index + brw->basevertex;
unsigned int max_index = brw->vb.max_index + brw->basevertex;
unsigned i;
int delta, j;
struct brw_vertex_element *upload[VERT_ATTRIB_MAX];
GLuint nr_uploads = 0;
/* _NEW_POLYGON
*
* On gen6+, edge flags don't end up in the VUE (either in or out of the
* VS). Instead, they're uploaded as the last vertex element, and the data
* is passed sideband through the fixed function units. So, we need to
* prepare the vertex buffer for it, but it's not present in inputs_read.
*/
if (brw->gen >= 6 && (ctx->Polygon.FrontMode != GL_FILL ||
ctx->Polygon.BackMode != GL_FILL)) {
vs_inputs |= VERT_BIT_EDGEFLAG;
}
if (0)
fprintf(stderr, "%s %d..%d\n", __func__, min_index, max_index);
/* Accumulate the list of enabled arrays. */
brw->vb.nr_enabled = 0;
while (vs_inputs) {
GLuint index = ffsll(vs_inputs) - 1;
struct brw_vertex_element *input = &brw->vb.inputs[index];
vs_inputs &= ~BITFIELD64_BIT(index);
brw->vb.enabled[brw->vb.nr_enabled++] = input;
}
if (brw->vb.nr_enabled == 0)
return;
if (brw->vb.nr_buffers)
return;
/* The range of data in a given buffer represented as [min, max) */
struct intel_buffer_object *enabled_buffer[VERT_ATTRIB_MAX];
uint32_t buffer_range_start[VERT_ATTRIB_MAX];
uint32_t buffer_range_end[VERT_ATTRIB_MAX];
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;
if (_mesa_is_bufferobj(glarray->BufferObj)) {
struct intel_buffer_object *intel_buffer =
intel_buffer_object(glarray->BufferObj);
const uint32_t offset = (uintptr_t)glarray->Ptr;
/* Start with the worst case */
uint32_t start = 0;
uint32_t range = intel_buffer->Base.Size;
if (glarray->InstanceDivisor) {
if (brw->num_instances) {
start = offset + glarray->StrideB * brw->baseinstance;
range = (glarray->StrideB * ((brw->num_instances - 1) /
glarray->InstanceDivisor) +
glarray->_ElementSize);
}
} else {
if (brw->vb.index_bounds_valid) {
start = offset + min_index * glarray->StrideB;
range = (glarray->StrideB * (max_index - min_index) +
glarray->_ElementSize);
}
}
/* If we have a VB set to be uploaded for this buffer object
* already, reuse that VB state so that we emit fewer
* relocations.
*/
unsigned 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;
buffer_range_start[input->buffer] =
MIN2(buffer_range_start[input->buffer], start);
buffer_range_end[input->buffer] =
MAX2(buffer_range_end[input->buffer], start + range);
break;
}
}
if (k == i) {
struct brw_vertex_buffer *buffer = &brw->vb.buffers[j];
/* Named buffer object: Just reference its contents directly. */
buffer->offset = offset;
buffer->stride = glarray->StrideB;
buffer->step_rate = glarray->InstanceDivisor;
buffer->size = glarray->BufferObj->Size - offset;
enabled_buffer[j] = intel_buffer;
buffer_range_start[j] = start;
buffer_range_end[j] = start + range;
input->buffer = j++;
input->offset = 0;
}
} 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) {
interleaved = glarray->StrideB;
ptr = glarray->Ptr;
}
else if (interleaved != glarray->StrideB ||
glarray->Ptr < ptr ||
(uintptr_t)(glarray->Ptr - ptr) + glarray->_ElementSize > interleaved)
{
/* If our stride is different from the first attribute's stride,
* or if the first attribute's stride didn't cover our element,
* disable the interleaved upload optimization. The second case
* can most commonly occur in cases where there is a single vertex
* and, for example, the data is stored on the application's
* stack.
*
* NOTE: This will also disable the optimization in cases where
* the data is in a different order than the array indices.
* Something like:
*
* float data[...];
* glVertexAttribPointer(0, 4, GL_FLOAT, 32, &data[4]);
* glVertexAttribPointer(1, 4, GL_FLOAT, 32, &data[0]);
*/
interleaved = 0;
}
upload[nr_uploads++] = input;
}
}
/* Now that we've set up all of the buffers, we walk through and reference
* each of them. We do this late so that we get the right size in each
* buffer and don't reference too little data.
*/
for (i = 0; i < j; i++) {
struct brw_vertex_buffer *buffer = &brw->vb.buffers[i];
if (buffer->bo)
continue;
const uint32_t start = buffer_range_start[i];
const uint32_t range = buffer_range_end[i] - buffer_range_start[i];
buffer->bo = intel_bufferobj_buffer(brw, enabled_buffer[i], start, range);
drm_intel_bo_reference(buffer->bo);
}
/* 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;
}
/* Handle any arrays to be uploaded. */
if (nr_uploads > 1) {
if (interleaved) {
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;
buffer->size += 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;
}
}
/* 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]->glarray->_ElementSize);
} 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]->glarray->_ElementSize);
}
buffer->offset -= delta * buffer->stride;
buffer->size += delta * buffer->stride;
buffer->step_rate = upload[i]->glarray->InstanceDivisor;
upload[i]->buffer = j++;
upload[i]->offset = 0;
}
brw->vb.nr_buffers = j;
}
