static void *r600_buffer_transfer_map(struct pipe_context *ctx,
struct pipe_resource *resource,
unsigned level,
unsigned usage,
const struct pipe_box *box,
struct pipe_transfer **ptransfer)
{
struct r600_context *rctx = (struct r600_context*)ctx;
struct r600_resource *rbuffer = r600_resource(resource);
uint8_t *data;
assert(box->x + box->width <= resource->width0);
if (usage & PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE &&
!(usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
assert(usage & PIPE_TRANSFER_WRITE);
/* Check if mapping this buffer would cause waiting for the GPU. */
if (rctx->ws->cs_is_buffer_referenced(rctx->cs, rbuffer->cs_buf, RADEON_USAGE_READWRITE) ||
rctx->ws->buffer_is_busy(rbuffer->buf, RADEON_USAGE_READWRITE)) {
unsigned i, mask;
/* Discard the buffer. */
pb_reference(&rbuffer->buf, NULL);
/* Create a new one in the same pipe_resource. */
/* XXX We probably want a different alignment for buffers and textures. */
r600_init_resource(rctx->screen, rbuffer, rbuffer->b.b.width0, 4096,
rbuffer->b.b.bind, rbuffer->b.b.usage);
/* We changed the buffer, now we need to bind it where the old one was bound. */
/* Vertex buffers. */
mask = rctx->vertex_buffer_state.enabled_mask;
while (mask) {
i = u_bit_scan(&mask);
if (rctx->vertex_buffer_state.vb[i].buffer == &rbuffer->b.b) {
rctx->vertex_buffer_state.dirty_mask |= 1 << i;
r600_vertex_buffers_dirty(rctx);
}
}
/* Streamout buffers. */
for (i = 0; i < rctx->num_so_targets; i++) {
if (rctx->so_targets[i]->b.buffer == &rbuffer->b.b) {
r600_context_streamout_end(rctx);
rctx->streamout_start = TRUE;
rctx->streamout_append_bitmask = ~0;
}
}
/* Constant buffers. */
r600_set_constants_dirty_if_bound(rctx, rbuffer);
}
}
else if ((usage & PIPE_TRANSFER_DISCARD_RANGE) &&
!(usage & PIPE_TRANSFER_UNSYNCHRONIZED) &&
rctx->screen->has_streamout &&
/* The buffer range must be aligned to 4. */
box->x % 4 == 0 && box->width % 4 == 0) {
assert(usage & PIPE_TRANSFER_WRITE);
/* Check if mapping this buffer would cause waiting for the GPU. */
if (rctx->ws->cs_is_buffer_referenced(rctx->cs, rbuffer->cs_buf, RADEON_USAGE_READWRITE) ||
rctx->ws->buffer_is_busy(rbuffer->buf, RADEON_USAGE_READWRITE)) {
/* Do a wait-free write-only transfer using a temporary buffer. */
struct r600_resource *staging = (struct r600_resource*)
pipe_buffer_create(ctx->screen, PIPE_BIND_VERTEX_BUFFER,
PIPE_USAGE_STAGING,
box->width + (box->x % R600_MAP_BUFFER_ALIGNMENT));
data = rctx->ws->buffer_map(staging->cs_buf, rctx->cs, PIPE_TRANSFER_WRITE);
if (!data)
return NULL;
data += box->x % R600_MAP_BUFFER_ALIGNMENT;
return r600_buffer_get_transfer(ctx, resource, level, usage, box,
ptransfer, data, staging);
}
}
data = rctx->ws->buffer_map(rbuffer->cs_buf, rctx->cs, usage);
if (!data) {
return NULL;
}
data += box->x;
return r600_buffer_get_transfer(ctx, resource, level, usage, box,
ptransfer, data, NULL);
}
static void *r600_buffer_transfer_map(struct pipe_context *ctx,
struct pipe_resource *resource,
unsigned level,
unsigned usage,
const struct pipe_box *box,
struct pipe_transfer **ptransfer)
{
struct r600_context *rctx = (struct r600_context*)ctx;
struct r600_resource *rbuffer = r600_resource(resource);
uint8_t *data;
assert(box->x + box->width <= resource->width0);
/* See if the buffer range being mapped has never been initialized,
* in which case it can be mapped unsynchronized. */
if (!(usage & PIPE_TRANSFER_UNSYNCHRONIZED) &&
usage & PIPE_TRANSFER_WRITE &&
!util_ranges_intersect(&rbuffer->valid_buffer_range, box->x, box->x + box->width)) {
usage |= PIPE_TRANSFER_UNSYNCHRONIZED;
}
if (usage & PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE &&
!(usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
assert(usage & PIPE_TRANSFER_WRITE);
/* Check if mapping this buffer would cause waiting for the GPU. */
if (r600_rings_is_buffer_referenced(&rctx->b, rbuffer->cs_buf, RADEON_USAGE_READWRITE) ||
rctx->b.ws->buffer_is_busy(rbuffer->buf, RADEON_USAGE_READWRITE)) {
unsigned i, mask;
/* Discard the buffer. */
pb_reference(&rbuffer->buf, NULL);
/* Create a new one in the same pipe_resource. */
/* XXX We probably want a different alignment for buffers and textures. */
r600_init_resource(&rctx->screen->b, rbuffer, rbuffer->b.b.width0, 4096,
TRUE, rbuffer->b.b.usage);
/* We changed the buffer, now we need to bind it where the old one was bound. */
/* Vertex buffers. */
mask = rctx->vertex_buffer_state.enabled_mask;
while (mask) {
i = u_bit_scan(&mask);
if (rctx->vertex_buffer_state.vb[i].buffer == &rbuffer->b.b) {
rctx->vertex_buffer_state.dirty_mask |= 1 << i;
r600_vertex_buffers_dirty(rctx);
}
}
/* Streamout buffers. */
for (i = 0; i < rctx->b.streamout.num_targets; i++) {
if (rctx->b.streamout.targets[i]->b.buffer == &rbuffer->b.b) {
if (rctx->b.streamout.begin_emitted) {
r600_emit_streamout_end(&rctx->b);
}
rctx->b.streamout.append_bitmask = rctx->b.streamout.enabled_mask;
r600_streamout_buffers_dirty(&rctx->b);
}
}
/* Constant buffers. */
r600_set_constants_dirty_if_bound(rctx, rbuffer);
}
}
else if ((usage & PIPE_TRANSFER_DISCARD_RANGE) &&
!(usage & PIPE_TRANSFER_UNSYNCHRONIZED) &&
!(rctx->screen->b.debug_flags & DBG_NO_DISCARD_RANGE) &&
(rctx->screen->has_cp_dma ||
(rctx->screen->has_streamout &&
/* The buffer range must be aligned to 4 with streamout. */
box->x % 4 == 0 && box->width % 4 == 0))) {
assert(usage & PIPE_TRANSFER_WRITE);
/* Check if mapping this buffer would cause waiting for the GPU. */
if (r600_rings_is_buffer_referenced(&rctx->b, rbuffer->cs_buf, RADEON_USAGE_READWRITE) ||
rctx->b.ws->buffer_is_busy(rbuffer->buf, RADEON_USAGE_READWRITE)) {
/* Do a wait-free write-only transfer using a temporary buffer. */
unsigned offset;
struct r600_resource *staging = NULL;
u_upload_alloc(rctx->uploader, 0, box->width + (box->x % R600_MAP_BUFFER_ALIGNMENT),
&offset, (struct pipe_resource**)&staging, (void**)&data);
if (staging) {
data += box->x % R600_MAP_BUFFER_ALIGNMENT;
return r600_buffer_get_transfer(ctx, resource, level, usage, box,
ptransfer, data, staging, offset);
}
}
}
/* mmap and synchronize with rings */
data = r600_buffer_map_sync_with_rings(&rctx->b, rbuffer, usage);
if (!data) {
return NULL;
}
data += box->x;
return r600_buffer_get_transfer(ctx, resource, level, usage, box,
ptransfer, data, NULL, 0);
}
static void *r600_buffer_transfer_map(struct pipe_context *ctx,
struct pipe_resource *resource,
unsigned level,
unsigned usage,
const struct pipe_box *box,
struct pipe_transfer **ptransfer)
{
struct r600_common_context *rctx = (struct r600_common_context*)ctx;
struct r600_common_screen *rscreen = (struct r600_common_screen*)ctx->screen;
struct r600_resource *rbuffer = r600_resource(resource);
uint8_t *data;
assert(box->x + box->width <= resource->width0);
/* See if the buffer range being mapped has never been initialized,
* in which case it can be mapped unsynchronized. */
if (!(usage & PIPE_TRANSFER_UNSYNCHRONIZED) &&
usage & PIPE_TRANSFER_WRITE &&
!util_ranges_intersect(&rbuffer->valid_buffer_range, box->x, box->x + box->width)) {
usage |= PIPE_TRANSFER_UNSYNCHRONIZED;
}
/* If discarding the entire range, discard the whole resource instead. */
if (usage & PIPE_TRANSFER_DISCARD_RANGE &&
box->x == 0 && box->width == resource->width0) {
usage |= PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE;
}
if (usage & PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE &&
!(usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
assert(usage & PIPE_TRANSFER_WRITE);
if (r600_invalidate_buffer(rctx, rbuffer)) {
/* At this point, the buffer is always idle. */
usage |= PIPE_TRANSFER_UNSYNCHRONIZED;
}
}
else if ((usage & PIPE_TRANSFER_DISCARD_RANGE) &&
!(usage & PIPE_TRANSFER_UNSYNCHRONIZED) &&
!(rscreen->debug_flags & DBG_NO_DISCARD_RANGE) &&
r600_can_dma_copy_buffer(rctx, box->x, 0, box->width)) {
assert(usage & PIPE_TRANSFER_WRITE);
/* Check if mapping this buffer would cause waiting for the GPU. */
if (r600_rings_is_buffer_referenced(rctx, rbuffer->buf, RADEON_USAGE_READWRITE) ||
!rctx->ws->buffer_wait(rbuffer->buf, 0, RADEON_USAGE_READWRITE)) {
/* Do a wait-free write-only transfer using a temporary buffer. */
unsigned offset;
struct r600_resource *staging = NULL;
u_upload_alloc(rctx->uploader, 0, box->width + (box->x % R600_MAP_BUFFER_ALIGNMENT),
256, &offset, (struct pipe_resource**)&staging, (void**)&data);
if (staging) {
data += box->x % R600_MAP_BUFFER_ALIGNMENT;
return r600_buffer_get_transfer(ctx, resource, level, usage, box,
ptransfer, data, staging, offset);
}
} else {
/* At this point, the buffer is always idle (we checked it above). */
usage |= PIPE_TRANSFER_UNSYNCHRONIZED;
}
}
/* Using a staging buffer in GTT for larger reads is much faster. */
else if ((usage & PIPE_TRANSFER_READ) &&
!(usage & PIPE_TRANSFER_WRITE) &&
rbuffer->domains == RADEON_DOMAIN_VRAM &&
r600_can_dma_copy_buffer(rctx, 0, box->x, box->width)) {
struct r600_resource *staging;
staging = (struct r600_resource*) pipe_buffer_create(
ctx->screen, PIPE_BIND_TRANSFER_READ, PIPE_USAGE_STAGING,
box->width + (box->x % R600_MAP_BUFFER_ALIGNMENT));
if (staging) {
/* Copy the VRAM buffer to the staging buffer. */
rctx->dma_copy(ctx, &staging->b.b, 0,
box->x % R600_MAP_BUFFER_ALIGNMENT,
0, 0, resource, level, box);
data = r600_buffer_map_sync_with_rings(rctx, staging, PIPE_TRANSFER_READ);
data += box->x % R600_MAP_BUFFER_ALIGNMENT;
return r600_buffer_get_transfer(ctx, resource, level, usage, box,
ptransfer, data, staging, 0);
}
}
data = r600_buffer_map_sync_with_rings(rctx, rbuffer, usage);
if (!data) {
return NULL;
}
data += box->x;
return r600_buffer_get_transfer(ctx, resource, level, usage, box,
ptransfer, data, NULL, 0);
}
static void *r600_buffer_transfer_map(struct pipe_context *ctx,
struct pipe_resource *resource,
unsigned level,
unsigned usage,
const struct pipe_box *box,
struct pipe_transfer **ptransfer)
{
struct r600_common_context *rctx = (struct r600_common_context*)ctx;
struct r600_common_screen *rscreen = (struct r600_common_screen*)ctx->screen;
struct r600_resource *rbuffer = r600_resource(resource);
uint8_t *data;
assert(box->x + box->width <= resource->width0);
/* See if the buffer range being mapped has never been initialized,
* in which case it can be mapped unsynchronized. */
if (!(usage & PIPE_TRANSFER_UNSYNCHRONIZED) &&
usage & PIPE_TRANSFER_WRITE &&
!util_ranges_intersect(&rbuffer->valid_buffer_range, box->x, box->x + box->width)) {
usage |= PIPE_TRANSFER_UNSYNCHRONIZED;
}
/* If discarding the entire range, discard the whole resource instead. */
if (usage & PIPE_TRANSFER_DISCARD_RANGE &&
box->x == 0 && box->width == resource->width0) {
usage |= PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE;
}
if (usage & PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE &&
!(usage & PIPE_TRANSFER_UNSYNCHRONIZED)) {
assert(usage & PIPE_TRANSFER_WRITE);
/* Check if mapping this buffer would cause waiting for the GPU. */
if (r600_rings_is_buffer_referenced(rctx, rbuffer->cs_buf, RADEON_USAGE_READWRITE) ||
rctx->ws->buffer_is_busy(rbuffer->buf, RADEON_USAGE_READWRITE)) {
rctx->invalidate_buffer(&rctx->b, &rbuffer->b.b);
}
/* At this point, the buffer is always idle. */
usage |= PIPE_TRANSFER_UNSYNCHRONIZED;
}
else if ((usage & PIPE_TRANSFER_DISCARD_RANGE) &&
!(usage & PIPE_TRANSFER_UNSYNCHRONIZED) &&
!(rscreen->debug_flags & DBG_NO_DISCARD_RANGE) &&
(rscreen->has_cp_dma ||
(rscreen->has_streamout &&
/* The buffer range must be aligned to 4 with streamout. */
box->x % 4 == 0 && box->width % 4 == 0))) {
assert(usage & PIPE_TRANSFER_WRITE);
/* Check if mapping this buffer would cause waiting for the GPU. */
if (r600_rings_is_buffer_referenced(rctx, rbuffer->cs_buf, RADEON_USAGE_READWRITE) ||
rctx->ws->buffer_is_busy(rbuffer->buf, RADEON_USAGE_READWRITE)) {
/* Do a wait-free write-only transfer using a temporary buffer. */
unsigned offset;
struct r600_resource *staging = NULL;
u_upload_alloc(rctx->uploader, 0, box->width + (box->x % R600_MAP_BUFFER_ALIGNMENT),
&offset, (struct pipe_resource**)&staging, (void**)&data);
if (staging) {
data += box->x % R600_MAP_BUFFER_ALIGNMENT;
return r600_buffer_get_transfer(ctx, resource, level, usage, box,
ptransfer, data, staging, offset);
} else {
return NULL; /* error, shouldn't occur though */
}
}
/* At this point, the buffer is always idle (we checked it above). */
usage |= PIPE_TRANSFER_UNSYNCHRONIZED;
}
data = r600_buffer_map_sync_with_rings(rctx, rbuffer, usage);
if (!data) {
return NULL;
}
data += box->x;
return r600_buffer_get_transfer(ctx, resource, level, usage, box,
ptransfer, data, NULL, 0);
}
