#include "svga_hw_reg.h"
+/**
+ * Return a new index buffer which contains a translation of the original
+ * index buffer. An example of a translation is converting from QUAD
+ * primitives to TRIANGLE primitives. Each set of four indexes for a quad
+ * will be converted to six indices for two triangles.
+ *
+ * Before generating the new index buffer we'll check if the incoming
+ * buffer already has a translated buffer that can be re-used.
+ * This benefits demos like Cinebench R15 which has many
+ * glDrawElements(GL_QUADS) commands (we can't draw quads natively).
+ *
+ * \param offset offset in bytes to first index to translate in src buffer
+ * \param orig_prim original primitive type (like PIPE_PRIM_QUADS)
+ * \param gen_prim new/generated primitive type (like PIPE_PRIM_TRIANGLES)
+ * \param orig_nr number of indexes to translate in source buffer
+ * \param gen_nr number of indexes to write into new/dest buffer
+ * \param index_size bytes per index (2 or 4)
+ * \param translate the translation function from the u_translate module
+ * \param out_buf returns the new/translated index buffer
+ * \return error code to indicate success failure
+ */
static enum pipe_error
-translate_indices(struct svga_hwtnl *hwtnl, struct pipe_resource *src,
- unsigned offset, unsigned prim, unsigned nr,
- unsigned index_size,
- u_translate_func translate, struct pipe_resource **out_buf)
+translate_indices(struct svga_hwtnl *hwtnl,
+ const struct pipe_draw_info *info,
+ enum pipe_prim_type gen_prim,
+ unsigned orig_nr, unsigned gen_nr,
+ unsigned gen_size,
+ u_translate_func translate,
+ struct pipe_resource **out_buf,
+ unsigned *out_offset)
{
struct pipe_context *pipe = &hwtnl->svga->pipe;
+ struct svga_screen *screen = svga_screen(pipe->screen);
+ struct svga_buffer *src_sbuf = NULL;
struct pipe_transfer *src_transfer = NULL;
struct pipe_transfer *dst_transfer = NULL;
- unsigned size = index_size * nr;
+ const unsigned size = gen_size * gen_nr;
+ const unsigned offset = info->start * info->index_size;
const void *src_map = NULL;
struct pipe_resource *dst = NULL;
void *dst_map = NULL;
+ assert(gen_size == 2 || gen_size == 4);
+ if (!info->has_user_indices)
+ src_sbuf = svga_buffer(info->index.resource);
+
+ /* If the draw_info provides us with a buffer rather than a
+ * user pointer, Check to see if we've already translated that buffer
+ */
+ if (src_sbuf && !screen->debug.no_cache_index_buffers) {
+ /* Check if we already have a translated index buffer */
+ if (src_sbuf->translated_indices.buffer &&
+ src_sbuf->translated_indices.orig_prim == info->mode &&
+ src_sbuf->translated_indices.new_prim == gen_prim &&
+ src_sbuf->translated_indices.offset == offset &&
+ src_sbuf->translated_indices.count == orig_nr &&
+ src_sbuf->translated_indices.index_size == gen_size) {
+ pipe_resource_reference(out_buf, src_sbuf->translated_indices.buffer);
+ return PIPE_OK;
+ }
+ }
+
/* Need to trim vertex count to make sure we don't write too much data
* to the dst buffer in the translate() call.
*/
- u_trim_pipe_prim(prim, &nr);
+ u_trim_pipe_prim(gen_prim, &gen_nr);
- size = index_size * nr;
+ if (src_sbuf) {
+ /* If we have a source buffer, create a destination buffer in the
+ * hope that we can reuse the translated data later. If not,
+ * we'd probably be better off using the upload buffer.
+ */
+ dst = pipe_buffer_create(pipe->screen,
+ PIPE_BIND_INDEX_BUFFER, PIPE_USAGE_IMMUTABLE,
+ size);
+ if (!dst)
+ goto fail;
- dst = pipe_buffer_create(pipe->screen,
- PIPE_BIND_INDEX_BUFFER, PIPE_USAGE_DEFAULT, size);
- if (dst == NULL)
- goto fail;
+ dst_map = pipe_buffer_map(pipe, dst, PIPE_TRANSFER_WRITE, &dst_transfer);
+ if (!dst_map)
+ goto fail;
- src_map = pipe_buffer_map(pipe, src, PIPE_TRANSFER_READ, &src_transfer);
- if (src_map == NULL)
- goto fail;
+ *out_offset = 0;
+ src_map = pipe_buffer_map(pipe, info->index.resource,
+ PIPE_TRANSFER_READ |
+ PIPE_TRANSFER_UNSYNCHRONIZED,
+ &src_transfer);
+ if (!src_map)
+ goto fail;
+ } else {
+ /* Allocate upload buffer space. Align to the index size. */
+ u_upload_alloc(pipe->stream_uploader, 0, size, gen_size,
+ out_offset, &dst, &dst_map);
+ if (!dst)
+ goto fail;
- dst_map = pipe_buffer_map(pipe, dst, PIPE_TRANSFER_WRITE, &dst_transfer);
- if (dst_map == NULL)
- goto fail;
+ src_map = info->index.user;
+ }
+
+ translate((const char *) src_map + offset, 0, 0, gen_nr, 0, dst_map);
- translate((const char *) src_map + offset, 0, 0, nr, 0, dst_map);
+ if (src_transfer)
+ pipe_buffer_unmap(pipe, src_transfer);
- pipe_buffer_unmap(pipe, src_transfer);
- pipe_buffer_unmap(pipe, dst_transfer);
+ if (dst_transfer)
+ pipe_buffer_unmap(pipe, dst_transfer);
+ else
+ u_upload_unmap(pipe->stream_uploader);
*out_buf = dst;
+
+ if (src_sbuf && !screen->debug.no_cache_index_buffers) {
+ /* Save the new, translated index buffer in the hope we can use it
+ * again in the future.
+ */
+ pipe_resource_reference(&src_sbuf->translated_indices.buffer, dst);
+ src_sbuf->translated_indices.orig_prim = info->mode;
+ src_sbuf->translated_indices.new_prim = gen_prim;
+ src_sbuf->translated_indices.offset = offset;
+ src_sbuf->translated_indices.count = orig_nr;
+ src_sbuf->translated_indices.index_size = gen_size;
+ }
+
return PIPE_OK;
fail:
- if (src_map)
+ if (src_transfer)
pipe_buffer_unmap(pipe, src_transfer);
- if (dst_map)
+ if (dst_transfer)
pipe_buffer_unmap(pipe, dst_transfer);
+ else if (dst_map)
+ u_upload_unmap(pipe->stream_uploader);
if (dst)
- pipe->screen->resource_destroy(pipe->screen, dst);
+ pipe_resource_reference(&dst, NULL);
return PIPE_ERROR_OUT_OF_MEMORY;
}
struct pipe_resource *index_buffer,
unsigned index_size, int index_bias,
unsigned min_index, unsigned max_index,
- unsigned prim, unsigned start,
+ enum pipe_prim_type prim, unsigned start,
unsigned count,
unsigned start_instance,
- unsigned instance_count)
+ unsigned instance_count,
+ ubyte vertices_per_patch)
{
SVGA3dPrimitiveRange range;
unsigned hw_prim;
unsigned hw_count;
unsigned index_offset = start * index_size;
- hw_prim = svga_translate_prim(prim, count, &hw_count);
+ hw_prim = svga_translate_prim(prim, count, &hw_count, vertices_per_patch);
if (hw_count == 0)
return PIPE_OK; /* nothing to draw */
return svga_hwtnl_prim(hwtnl, &range, count,
min_index, max_index, index_buffer,
- start_instance, instance_count);
+ start_instance, instance_count,
+ NULL, NULL);
}
enum pipe_error
svga_hwtnl_draw_range_elements(struct svga_hwtnl *hwtnl,
- struct pipe_resource *index_buffer,
- unsigned index_size, int index_bias,
- unsigned min_index, unsigned max_index,
- unsigned prim, unsigned start, unsigned count,
- unsigned start_instance, unsigned instance_count)
+ const struct pipe_draw_info *info,
+ unsigned count)
{
- unsigned gen_prim, gen_size, gen_nr;
+ struct pipe_context *pipe = &hwtnl->svga->pipe;
+ enum pipe_prim_type gen_prim;
+ unsigned gen_size, gen_nr;
enum indices_mode gen_type;
u_translate_func gen_func;
enum pipe_error ret = PIPE_OK;
- if (hwtnl->api_fillmode != PIPE_POLYGON_MODE_FILL &&
- prim >= PIPE_PRIM_TRIANGLES) {
- gen_type = u_unfilled_translator(prim,
- index_size,
+ SVGA_STATS_TIME_PUSH(svga_sws(hwtnl->svga),
+ SVGA_STATS_TIME_HWTNLDRAWELEMENTS);
+
+ if (svga_need_unfilled_fallback(hwtnl, info->mode)) {
+ gen_type = u_unfilled_translator(info->mode,
+ info->index_size,
count,
hwtnl->api_fillmode,
&gen_prim,
&gen_size, &gen_nr, &gen_func);
}
else {
+ unsigned hw_pv;
+
+ /* There is no geometry ordering with PATCH, so no need to
+ * consider provoking vertex mode for the translation.
+ * So use the same api_pv as the hw_pv.
+ */
+ hw_pv = info->mode == PIPE_PRIM_PATCHES ? hwtnl->api_pv :
+ hwtnl->hw_pv;
gen_type = u_index_translator(svga_hw_prims,
- prim,
- index_size,
+ info->mode,
+ info->index_size,
count,
hwtnl->api_pv,
- hwtnl->hw_pv,
+ hw_pv,
PR_DISABLE,
&gen_prim, &gen_size, &gen_nr, &gen_func);
}
if (gen_type == U_TRANSLATE_MEMCPY) {
/* No need for translation, just pass through to hardware:
*/
- return svga_hwtnl_simple_draw_range_elements(hwtnl, index_buffer,
- index_size,
- index_bias,
- min_index,
- max_index,
- gen_prim, start, count,
- start_instance,
- instance_count);
+ unsigned start_offset = info->start * info->index_size;
+ struct pipe_resource *index_buffer = NULL;
+ unsigned index_offset;
+
+ if (info->has_user_indices) {
+ u_upload_data(pipe->stream_uploader, 0, count * info->index_size,
+ info->index_size, (char *) info->index.user + start_offset,
+ &index_offset, &index_buffer);
+ u_upload_unmap(pipe->stream_uploader);
+ index_offset /= info->index_size;
+ } else {
+ pipe_resource_reference(&index_buffer, info->index.resource);
+ index_offset = info->start;
+ }
+
+ assert(index_buffer != NULL);
+
+ ret = svga_hwtnl_simple_draw_range_elements(hwtnl, index_buffer,
+ info->index_size,
+ info->index_bias,
+ info->min_index,
+ info->max_index,
+ gen_prim, index_offset, count,
+ info->start_instance,
+ info->instance_count,
+ info->vertices_per_patch);
+ pipe_resource_reference(&index_buffer, NULL);
}
else {
struct pipe_resource *gen_buf = NULL;
+ unsigned gen_offset = 0;
/* Need to allocate a new index buffer and run the translate
* func to populate it. Could potentially cache this translated
* GL though, as index buffers are typically used only once
* there.
*/
- ret = translate_indices(hwtnl,
- index_buffer,
- start * index_size,
- gen_prim, gen_nr, gen_size, gen_func, &gen_buf);
- if (ret != PIPE_OK)
- goto done;
-
- ret = svga_hwtnl_simple_draw_range_elements(hwtnl,
- gen_buf,
- gen_size,
- index_bias,
- min_index,
- max_index,
- gen_prim, 0, gen_nr,
- start_instance,
- instance_count);
- if (ret != PIPE_OK)
- goto done;
-
-done:
- if (gen_buf)
- pipe_resource_reference(&gen_buf, NULL);
+ ret = translate_indices(hwtnl, info, gen_prim,
+ count, gen_nr, gen_size,
+ gen_func, &gen_buf, &gen_offset);
+ if (ret == PIPE_OK) {
+ gen_offset /= gen_size;
+ ret = svga_hwtnl_simple_draw_range_elements(hwtnl,
+ gen_buf,
+ gen_size,
+ info->index_bias,
+ info->min_index,
+ info->max_index,
+ gen_prim, gen_offset,
+ gen_nr,
+ info->start_instance,
+ info->instance_count,
+ info->vertices_per_patch);
+ }
- return ret;
+ if (gen_buf) {
+ pipe_resource_reference(&gen_buf, NULL);
+ }
}
+
+ SVGA_STATS_TIME_POP(svga_sws(hwtnl->svga));
+ return ret;
}
projects / mesa.git / blobdiff