summaryrefslogtreecommitdiff
path: root/src/gallium/auxiliary/draw/draw_pt_vcache.c
diff options
context:
space:
mode:
Diffstat (limited to 'src/gallium/auxiliary/draw/draw_pt_vcache.c')
-rw-r--r--src/gallium/auxiliary/draw/draw_pt_vcache.c124
1 files changed, 78 insertions, 46 deletions
diff --git a/src/gallium/auxiliary/draw/draw_pt_vcache.c b/src/gallium/auxiliary/draw/draw_pt_vcache.c
index 914c87a9dc..8ef94c3163 100644
--- a/src/gallium/auxiliary/draw/draw_pt_vcache.c
+++ b/src/gallium/auxiliary/draw/draw_pt_vcache.c
@@ -41,6 +41,7 @@
#define FETCH_MAX 256
#define DRAW_MAX (16*1024)
+
struct vcache_frontend {
struct draw_pt_front_end base;
struct draw_context *draw;
@@ -64,13 +65,13 @@ struct vcache_frontend {
unsigned opt;
};
+
static INLINE void
vcache_flush( struct vcache_frontend *vcache )
{
if (vcache->middle_prim != vcache->output_prim) {
vcache->middle_prim = vcache->output_prim;
vcache->middle->prepare( vcache->middle,
- vcache->input_prim,
vcache->middle_prim,
vcache->opt,
&vcache->fetch_max );
@@ -89,12 +90,12 @@ vcache_flush( struct vcache_frontend *vcache )
vcache->draw_count = 0;
}
+
static INLINE void
vcache_check_flush( struct vcache_frontend *vcache )
{
- if ( vcache->draw_count + 6 >= DRAW_MAX ||
- vcache->fetch_count + 4 >= FETCH_MAX )
- {
+ if (vcache->draw_count + 6 >= DRAW_MAX ||
+ vcache->fetch_count + 4 >= FETCH_MAX) {
vcache_flush( vcache );
}
}
@@ -146,6 +147,7 @@ vcache_triangle_flags( struct vcache_frontend *vcache,
vcache_check_flush(vcache);
}
+
static INLINE void
vcache_line( struct vcache_frontend *vcache,
unsigned i0,
@@ -177,6 +179,7 @@ vcache_point( struct vcache_frontend *vcache,
vcache_check_flush(vcache);
}
+
static INLINE void
vcache_quad( struct vcache_frontend *vcache,
unsigned i0,
@@ -196,6 +199,7 @@ vcache_quad( struct vcache_frontend *vcache,
}
}
+
static INLINE void
vcache_ef_quad( struct vcache_frontend *vcache,
unsigned i0,
@@ -231,6 +235,7 @@ vcache_ef_quad( struct vcache_frontend *vcache,
}
}
+
/* At least for now, we're back to using a template include file for
* this. The two paths aren't too different though - it may be
* possible to reunify them.
@@ -256,23 +261,23 @@ rebase_uint_elts( const unsigned *src,
ushort *dest )
{
unsigned i;
-
for (i = 0; i < count; i++)
dest[i] = (ushort)(src[i] + delta);
}
+
static INLINE void
rebase_ushort_elts( const ushort *src,
unsigned count,
int delta,
- ushort *dest )
+ ushort *dest )
{
unsigned i;
-
for (i = 0; i < count; i++)
dest[i] = (ushort)(src[i] + delta);
}
+
static INLINE void
rebase_ubyte_elts( const ubyte *src,
unsigned count,
@@ -280,42 +285,39 @@ rebase_ubyte_elts( const ubyte *src,
ushort *dest )
{
unsigned i;
-
for (i = 0; i < count; i++)
dest[i] = (ushort)(src[i] + delta);
}
-
static INLINE void
translate_uint_elts( const unsigned *src,
unsigned count,
ushort *dest )
{
unsigned i;
-
for (i = 0; i < count; i++)
dest[i] = (ushort)(src[i]);
}
+
static INLINE void
translate_ushort_elts( const ushort *src,
unsigned count,
ushort *dest )
{
unsigned i;
-
for (i = 0; i < count; i++)
dest[i] = (ushort)(src[i]);
}
+
static INLINE void
translate_ubyte_elts( const ubyte *src,
unsigned count,
ushort *dest )
{
unsigned i;
-
for (i = 0; i < count; i++)
dest[i] = (ushort)(src[i]);
}
@@ -336,6 +338,7 @@ format_from_get_elt( pt_elt_func get_elt )
}
#endif
+
static INLINE void
vcache_check_run( struct draw_pt_front_end *frontend,
pt_elt_func get_elt,
@@ -345,18 +348,46 @@ vcache_check_run( struct draw_pt_front_end *frontend,
{
struct vcache_frontend *vcache = (struct vcache_frontend *)frontend;
struct draw_context *draw = vcache->draw;
- unsigned min_index = draw->pt.user.min_index;
- unsigned max_index = draw->pt.user.max_index;
- unsigned index_size = draw->pt.user.eltSize;
- unsigned fetch_count = max_index + 1 - min_index;
+ const unsigned min_index = draw->pt.user.min_index;
+ const unsigned max_index = draw->pt.user.max_index;
+ const unsigned index_size = draw->pt.user.eltSize;
+ unsigned fetch_count;
const ushort *transformed_elts;
ushort *storage = NULL;
boolean ok = FALSE;
+ /* debug: verify indexes are in range [min_index, max_index] */
+ if (0) {
+ unsigned i;
+ for (i = 0; i < draw_count; i++) {
+ if (index_size == 1) {
+ assert( ((const ubyte *) elts)[i] >= min_index);
+ assert( ((const ubyte *) elts)[i] <= max_index);
+ }
+ else if (index_size == 2) {
+ assert( ((const ushort *) elts)[i] >= min_index);
+ assert( ((const ushort *) elts)[i] <= max_index);
+ }
+ else {
+ assert(index_size == 4);
+ assert( ((const uint *) elts)[i] >= min_index);
+ assert( ((const uint *) elts)[i] <= max_index);
+ }
+ }
+ }
+
+ /* Note: max_index is frequently 0xffffffff so we have to be sure
+ * that any arithmetic involving max_index doesn't overflow!
+ */
+ if (max_index >= (unsigned) DRAW_PIPE_MAX_VERTICES)
+ goto fail;
- if (0) debug_printf("fetch_count %d fetch_max %d draw_count %d\n", fetch_count,
- vcache->fetch_max,
- draw_count);
+ fetch_count = max_index + 1 - min_index;
+
+ if (0)
+ debug_printf("fetch_count %d fetch_max %d draw_count %d\n", fetch_count,
+ vcache->fetch_max,
+ draw_count);
if (elt_bias + max_index >= DRAW_PIPE_MAX_VERTICES ||
fetch_count >= UNDEFINED_VERTEX_ID ||
@@ -368,23 +399,19 @@ vcache_check_run( struct draw_pt_front_end *frontend,
if (vcache->middle_prim != vcache->input_prim) {
vcache->middle_prim = vcache->input_prim;
vcache->middle->prepare( vcache->middle,
- vcache->input_prim,
vcache->middle_prim,
vcache->opt,
&vcache->fetch_max );
}
-
assert((elt_bias >= 0 && min_index + elt_bias >= min_index) ||
(elt_bias < 0 && min_index + elt_bias < min_index));
if (min_index == 0 &&
- index_size == 2)
- {
+ index_size == 2) {
transformed_elts = (const ushort *)elts;
}
- else
- {
+ else {
storage = MALLOC( draw_count * sizeof(ushort) );
if (!storage)
goto fail;
@@ -419,23 +446,23 @@ vcache_check_run( struct draw_pt_front_end *frontend,
switch(index_size) {
case 1:
rebase_ubyte_elts( (const ubyte *)elts,
- draw_count,
- 0 - (int)min_index,
- storage );
+ draw_count,
+ 0 - (int)min_index,
+ storage );
break;
case 2:
rebase_ushort_elts( (const ushort *)elts,
- draw_count,
- 0 - (int)min_index,
- storage );
+ draw_count,
+ 0 - (int)min_index,
+ storage );
break;
case 4:
rebase_uint_elts( (const uint *)elts,
- draw_count,
- 0 - (int)min_index,
- storage );
+ draw_count,
+ 0 - (int)min_index,
+ storage );
break;
default:
@@ -462,7 +489,7 @@ vcache_check_run( struct draw_pt_front_end *frontend,
debug_printf("failed to execute atomic draw elts for %d/%d, splitting up\n",
fetch_count, draw_count);
- fail:
+fail:
vcache_run( frontend, get_elt, elts, elt_bias, draw_count );
}
@@ -472,23 +499,26 @@ vcache_check_run( struct draw_pt_front_end *frontend,
static void
vcache_prepare( struct draw_pt_front_end *frontend,
unsigned in_prim,
- unsigned out_prim,
struct draw_pt_middle_end *middle,
unsigned opt )
{
struct vcache_frontend *vcache = (struct vcache_frontend *)frontend;
- if (opt & PT_PIPELINE)
- {
+ if (opt & PT_PIPELINE) {
vcache->base.run = vcache_run_extras;
}
- else
- {
+ else {
vcache->base.run = vcache_check_run;
}
+ /* VCache will always emit the reduced version of its input
+ * primitive, ie STRIP/FANS become TRIS, etc.
+ *
+ * This is not to be confused with what the GS might be up to,
+ * which is a separate issue.
+ */
vcache->input_prim = in_prim;
- vcache->output_prim = u_reduced_prim(out_prim);
+ vcache->output_prim = u_reduced_prim(in_prim);
vcache->middle = middle;
vcache->opt = opt;
@@ -496,12 +526,13 @@ vcache_prepare( struct draw_pt_front_end *frontend,
/* Have to run prepare here, but try and guess a good prim for
* doing so:
*/
- vcache->middle_prim = (opt & PT_PIPELINE) ? vcache->output_prim : vcache->input_prim;
- middle->prepare( middle, vcache->input_prim,
- vcache->middle_prim, opt, &vcache->fetch_max );
-}
-
+ vcache->middle_prim = (opt & PT_PIPELINE)
+ ? vcache->output_prim : vcache->input_prim;
+ middle->prepare( middle,
+ vcache->middle_prim,
+ opt, &vcache->fetch_max );
+}
static void
@@ -512,6 +543,7 @@ vcache_finish( struct draw_pt_front_end *frontend )
vcache->middle = NULL;
}
+
static void
vcache_destroy( struct draw_pt_front_end *frontend )
{
generated by cgit v1.2.3 (git 2.39.1) at 2025-04-14 15:23:20 +0000