summaryrefslogtreecommitdiffstats
path: root/src/gallium/drivers
diff options
context:
space:
mode:
authorAlyssa Rosenzweig <[email protected]>2020-02-19 10:15:13 -0500
committerMarge Bot <[email protected]>2020-02-27 10:30:48 +0000
commit12db69aa3f6155e9ccb1d783da589ab206dc7239 (patch)
tree8571927a6aa479df9c0f34f7a277dda3c56bacdf /src/gallium/drivers
parente272b110bb511b3735672a61c58c5b6b944bc37b (diff)
panfrost: Combine get_index_buffer with bound computation
These operations are intertwined since there are optimizations that will want to "double dip". In particular for user index buffers we'd want to upload simultaneous with index computation. For resources we'd like to keep resource related code together. Signed-off-by: Alyssa Rosenzweig <[email protected]> Reviewed-by: Boris Brezillon <[email protected]> Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3880>
Diffstat (limited to 'src/gallium/drivers')
-rw-r--r--src/gallium/drivers/panfrost/pan_context.c41
1 files changed, 25 insertions, 16 deletions
diff --git a/src/gallium/drivers/panfrost/pan_context.c b/src/gallium/drivers/panfrost/pan_context.c
index ff00c2129bf..0381e410437 100644
--- a/src/gallium/drivers/panfrost/pan_context.c
+++ b/src/gallium/drivers/panfrost/pan_context.c
@@ -1252,15 +1252,27 @@ panfrost_translate_index_size(unsigned size)
}
/* Gets a GPU address for the associated index buffer. Only gauranteed to be
- * good for the duration of the draw (transient), could last longer */
+ * good for the duration of the draw (transient), could last longer. Also get
+ * the bounds on the index buffer for the range accessed by the draw. We do
+ * these operations together because there are natural optimizations which
+ * require them to be together. */
static mali_ptr
-panfrost_get_index_buffer_mapped(struct panfrost_context *ctx, const struct pipe_draw_info *info)
+panfrost_get_index_buffer_bounded(struct panfrost_context *ctx, const struct pipe_draw_info *info, unsigned *min_index, unsigned *max_index)
{
struct panfrost_resource *rsrc = (struct panfrost_resource *) (info->index.resource);
off_t offset = info->start * info->index_size;
struct panfrost_batch *batch = panfrost_get_batch_for_fbo(ctx);
+ mali_ptr out = 0;
+
+ bool needs_indices = true;
+
+ if (info->max_index != ~0u) {
+ *min_index = info->min_index;
+ *max_index = info->max_index;
+ needs_indices = false;
+ }
if (!info->has_user_indices) {
/* Only resources can be directly mapped */
@@ -1268,12 +1280,19 @@ panfrost_get_index_buffer_mapped(struct panfrost_context *ctx, const struct pipe
PAN_BO_ACCESS_SHARED |
PAN_BO_ACCESS_READ |
PAN_BO_ACCESS_VERTEX_TILER);
- return rsrc->bo->gpu + offset;
+ out = rsrc->bo->gpu + offset;
} else {
/* Otherwise, we need to upload to transient memory */
const uint8_t *ibuf8 = (const uint8_t *) info->index.user;
- return panfrost_upload_transient(batch, ibuf8 + offset, info->count * info->index_size);
+ out = panfrost_upload_transient(batch, ibuf8 + offset, info->count * info->index_size);
+ }
+
+ if (needs_indices) {
+ /* Fallback */
+ u_vbuf_get_minmax_index(&ctx->base, info, min_index, max_index);
}
+
+ return out;
}
static bool
@@ -1391,18 +1410,9 @@ panfrost_draw_vbo(
panfrost_statistics_record(ctx, info);
if (info->index_size) {
- /* Calculate the min/max index used so we can figure out how
- * many times to invoke the vertex shader */
-
- /* Fetch / calculate index bounds */
unsigned min_index = 0, max_index = 0;
-
- if (info->max_index == ~0u) {
- u_vbuf_get_minmax_index(pipe, info, &min_index, &max_index);
- } else {
- min_index = info->min_index;
- max_index = info->max_index;
- }
+ ctx->payloads[PIPE_SHADER_FRAGMENT].prefix.indices =
+ panfrost_get_index_buffer_bounded(ctx, info, &min_index, &max_index);
/* Use the corresponding values */
vertex_count = max_index - min_index + 1;
@@ -1413,7 +1423,6 @@ panfrost_draw_vbo(
ctx->payloads[PIPE_SHADER_FRAGMENT].prefix.index_count = MALI_POSITIVE(info->count);
draw_flags |= panfrost_translate_index_size(info->index_size);
- ctx->payloads[PIPE_SHADER_FRAGMENT].prefix.indices = panfrost_get_index_buffer_mapped(ctx, info);
} else {
/* Index count == vertex count, if no indexing is applied, as
* if it is internally indexed in the expected order */