Mesa (master): r300g: implement instanced arrays
Marek Olšák
mareko at kemper.freedesktop.org
Sat Mar 5 18:34:51 UTC 2011
Module: Mesa
Branch: master
Commit: 65482f2c2b1c3456d0ca58a38d82c02a5d6d362c
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=65482f2c2b1c3456d0ca58a38d82c02a5d6d362c
Author: Marek Olšák <maraeo at gmail.com>
Date: Sat Mar 5 15:53:46 2011 +0100
r300g: implement instanced arrays
---
src/gallium/drivers/r300/r300_context.h | 2 +
src/gallium/drivers/r300/r300_emit.c | 100 ++++++++++++++++++++++++-------
src/gallium/drivers/r300/r300_emit.h | 3 +-
src/gallium/drivers/r300/r300_render.c | 73 +++++++++++++++-------
src/gallium/drivers/r300/r300_screen.c | 2 +-
5 files changed, 132 insertions(+), 48 deletions(-)
diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h
index 58e1094..1e28221 100644
--- a/src/gallium/drivers/r300/r300_context.h
+++ b/src/gallium/drivers/r300/r300_context.h
@@ -610,6 +610,8 @@ struct r300_context {
boolean vertex_arrays_dirty;
boolean vertex_arrays_indexed;
int vertex_arrays_offset;
+ int vertex_arrays_instance_id;
+ boolean instancing_enabled;
};
#define foreach_atom(r300, atom) \
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index e3945b7..60f8305 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -794,7 +794,8 @@ void r300_emit_textures_state(struct r300_context *r300,
END_CS;
}
-void r300_emit_vertex_arrays(struct r300_context* r300, int offset, boolean indexed)
+void r300_emit_vertex_arrays(struct r300_context* r300, int offset,
+ boolean indexed, int instance_id)
{
struct pipe_vertex_buffer *vbuf = r300->vbuf_mgr->vertex_buffer;
struct pipe_resource **valid_vbuf = r300->vbuf_mgr->real_vertex_buffer;
@@ -804,39 +805,92 @@ void r300_emit_vertex_arrays(struct r300_context* r300, int offset, boolean inde
unsigned vertex_array_count = r300->velems->count;
unsigned packet_size = (vertex_array_count * 3 + 1) / 2;
struct pipe_vertex_buffer *vb1, *vb2;
- unsigned *hw_format_size;
- unsigned size1, size2;
+ unsigned *hw_format_size = r300->velems->format_size;
+ unsigned size1, size2, offset1, offset2, stride1, stride2;
CS_LOCALS(r300);
BEGIN_CS(2 + packet_size + vertex_array_count * 2);
OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, packet_size);
OUT_CS(vertex_array_count | (!indexed ? R300_VC_FORCE_PREFETCH : 0));
- hw_format_size = r300->velems->format_size;
+ if (instance_id == -1) {
+ /* Non-instanced arrays. This ignores instance_divisor and instance_id. */
+ for (i = 0; i < vertex_array_count - 1; i += 2) {
+ vb1 = &vbuf[velem[i].vertex_buffer_index];
+ vb2 = &vbuf[velem[i+1].vertex_buffer_index];
+ size1 = hw_format_size[i];
+ size2 = hw_format_size[i+1];
+
+ OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride) |
+ R300_VBPNTR_SIZE1(size2) | R300_VBPNTR_STRIDE1(vb2->stride));
+ OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride);
+ OUT_CS(vb2->buffer_offset + velem[i+1].src_offset + offset * vb2->stride);
+ }
- for (i = 0; i < vertex_array_count - 1; i += 2) {
- vb1 = &vbuf[velem[i].vertex_buffer_index];
- vb2 = &vbuf[velem[i+1].vertex_buffer_index];
- size1 = hw_format_size[i];
- size2 = hw_format_size[i+1];
+ if (vertex_array_count & 1) {
+ vb1 = &vbuf[velem[i].vertex_buffer_index];
+ size1 = hw_format_size[i];
- OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride) |
- R300_VBPNTR_SIZE1(size2) | R300_VBPNTR_STRIDE1(vb2->stride));
- OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride);
- OUT_CS(vb2->buffer_offset + velem[i+1].src_offset + offset * vb2->stride);
- }
+ OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride));
+ OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride);
+ }
+
+ for (i = 0; i < vertex_array_count; i++) {
+ buf = r300_resource(valid_vbuf[velem[i].vertex_buffer_index]);
+ OUT_CS_RELOC(buf);
+ }
+ } else {
+ /* Instanced arrays. */
+ for (i = 0; i < vertex_array_count - 1; i += 2) {
+ vb1 = &vbuf[velem[i].vertex_buffer_index];
+ vb2 = &vbuf[velem[i+1].vertex_buffer_index];
+ size1 = hw_format_size[i];
+ size2 = hw_format_size[i+1];
+
+ if (velem[i].instance_divisor) {
+ stride1 = 0;
+ offset1 = vb1->buffer_offset + velem[i].src_offset +
+ (instance_id / velem[i].instance_divisor) * vb1->stride;
+ } else {
+ stride1 = vb1->stride;
+ offset1 = vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride;
+ }
+ if (velem[i+1].instance_divisor) {
+ stride2 = 0;
+ offset2 = vb2->buffer_offset + velem[i+1].src_offset +
+ (instance_id / velem[i+1].instance_divisor) * vb2->stride;
+ } else {
+ stride2 = vb2->stride;
+ offset2 = vb2->buffer_offset + velem[i+1].src_offset + offset * vb2->stride;
+ }
- if (vertex_array_count & 1) {
- vb1 = &vbuf[velem[i].vertex_buffer_index];
- size1 = hw_format_size[i];
+ OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(stride1) |
+ R300_VBPNTR_SIZE1(size2) | R300_VBPNTR_STRIDE1(stride2));
+ OUT_CS(offset1);
+ OUT_CS(offset2);
+ }
- OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(vb1->stride));
- OUT_CS(vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride);
- }
+ if (vertex_array_count & 1) {
+ vb1 = &vbuf[velem[i].vertex_buffer_index];
+ size1 = hw_format_size[i];
+
+ if (velem[i].instance_divisor) {
+ stride1 = 0;
+ offset1 = vb1->buffer_offset + velem[i].src_offset +
+ (instance_id / velem[i].instance_divisor) * vb1->stride;
+ } else {
+ stride1 = vb1->stride;
+ offset1 = vb1->buffer_offset + velem[i].src_offset + offset * vb1->stride;
+ }
- for (i = 0; i < vertex_array_count; i++) {
- buf = r300_resource(valid_vbuf[velem[i].vertex_buffer_index]);
- OUT_CS_RELOC(buf);
+ OUT_CS(R300_VBPNTR_SIZE0(size1) | R300_VBPNTR_STRIDE0(stride1));
+ OUT_CS(offset1);
+ }
+
+ for (i = 0; i < vertex_array_count; i++) {
+ buf = r300_resource(valid_vbuf[velem[i].vertex_buffer_index]);
+ OUT_CS_RELOC(buf);
+ }
}
END_CS;
}
diff --git a/src/gallium/drivers/r300/r300_emit.h b/src/gallium/drivers/r300/r300_emit.h
index acea51d..6c1c9d2 100644
--- a/src/gallium/drivers/r300/r300_emit.h
+++ b/src/gallium/drivers/r300/r300_emit.h
@@ -31,7 +31,8 @@ struct r300_vertex_program_code;
uint32_t pack_float24(float f);
-void r300_emit_vertex_arrays(struct r300_context* r300, int offset, boolean indexed);
+void r300_emit_vertex_arrays(struct r300_context* r300, int offset,
+ boolean indexed, int instance_id);
void r300_emit_blend_state(struct r300_context* r300,
unsigned size, void* state);
diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c
index c402a83..300cb86 100644
--- a/src/gallium/drivers/r300/r300_render.c
+++ b/src/gallium/drivers/r300/r300_render.c
@@ -233,13 +233,14 @@ static boolean r300_reserve_cs_dwords(struct r300_context *r300,
* \param index_buffer The index buffer to validate. The parameter may be NULL.
* \param buffer_offset The offset passed to emit_vertex_arrays.
* \param index_bias The index bias to emit.
+ * \param instance_id Index of instance to render
* \return TRUE if rendering should be skipped
*/
static boolean r300_emit_states(struct r300_context *r300,
enum r300_prepare_flags flags,
struct pipe_resource *index_buffer,
int buffer_offset,
- int index_bias)
+ int index_bias, int instance_id)
{
boolean first_draw = flags & PREP_EMIT_STATES;
boolean emit_vertex_arrays = flags & PREP_EMIT_AOS;
@@ -267,12 +268,14 @@ static boolean r300_emit_states(struct r300_context *r300,
if (emit_vertex_arrays &&
(r300->vertex_arrays_dirty ||
r300->vertex_arrays_indexed != indexed ||
- r300->vertex_arrays_offset != buffer_offset)) {
- r300_emit_vertex_arrays(r300, buffer_offset, indexed);
+ r300->vertex_arrays_offset != buffer_offset ||
+ r300->vertex_arrays_instance_id != instance_id)) {
+ r300_emit_vertex_arrays(r300, buffer_offset, indexed, instance_id);
r300->vertex_arrays_dirty = FALSE;
r300->vertex_arrays_indexed = indexed;
r300->vertex_arrays_offset = buffer_offset;
+ r300->vertex_arrays_instance_id = instance_id;
}
if (emit_vertex_arrays_swtcl)
@@ -291,6 +294,7 @@ static boolean r300_emit_states(struct r300_context *r300,
* \param cs_dwords The number of dwords to reserve in CS.
* \param buffer_offset The offset passed to emit_vertex_arrays.
* \param index_bias The index bias to emit.
+ * \param instance_id The instance to render.
* \return TRUE if rendering should be skipped
*/
static boolean r300_prepare_for_rendering(struct r300_context *r300,
@@ -298,14 +302,15 @@ static boolean r300_prepare_for_rendering(struct r300_context *r300,
struct pipe_resource *index_buffer,
unsigned cs_dwords,
int buffer_offset,
- int index_bias)
+ int index_bias,
+ int instance_id)
{
/* Make sure there is enough space in the command stream and emit states. */
if (r300_reserve_cs_dwords(r300, flags, cs_dwords))
flags |= PREP_EMIT_STATES;
return r300_emit_states(r300, flags, index_buffer, buffer_offset,
- index_bias);
+ index_bias, instance_id);
}
static boolean immd_is_good_idea(struct r300_context *r300,
@@ -379,7 +384,7 @@ static void r300_draw_arrays_immediate(struct r300_context *r300,
CS_LOCALS(r300);
- if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0))
+ if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1))
return;
/* Calculate the vertex size, offsets, strides etc. and map the buffers. */
@@ -540,7 +545,7 @@ static void r300_draw_elements_immediate(struct r300_context *r300,
/* 19 dwords for r300_draw_elements_immediate. Give up if the function fails. */
if (!r300_prepare_for_rendering(r300,
PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_AOS |
- PREP_INDEXED, NULL, 2+count_dwords, 0, info->index_bias))
+ PREP_INDEXED, NULL, 2+count_dwords, 0, info->index_bias, -1))
return;
r300_emit_draw_init(r300, info->mode, info->min_index, info->max_index);
@@ -612,7 +617,8 @@ static void r300_draw_elements_immediate(struct r300_context *r300,
}
static void r300_draw_elements(struct r300_context *r300,
- const struct pipe_draw_info *info)
+ const struct pipe_draw_info *info,
+ int instance_id)
{
struct pipe_resource *indexBuffer = r300->index_buffer.buffer;
unsigned indexSize = r300->index_buffer.index_size;
@@ -661,7 +667,8 @@ static void r300_draw_elements(struct r300_context *r300,
/* 19 dwords for emit_draw_elements. Give up if the function fails. */
if (!r300_prepare_for_rendering(r300,
PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_AOS |
- PREP_INDEXED, indexBuffer, 19, buffer_offset, info->index_bias))
+ PREP_INDEXED, indexBuffer, 19, buffer_offset, info->index_bias,
+ instance_id))
goto done;
if (alt_num_verts || count <= 65535) {
@@ -686,7 +693,8 @@ static void r300_draw_elements(struct r300_context *r300,
if (count) {
if (!r300_prepare_for_rendering(r300,
PREP_VALIDATE_VBOS | PREP_EMIT_AOS | PREP_INDEXED,
- indexBuffer, 19, buffer_offset, info->index_bias))
+ indexBuffer, 19, buffer_offset, info->index_bias,
+ instance_id))
goto done;
}
} while (count);
@@ -699,7 +707,8 @@ done:
}
static void r300_draw_arrays(struct r300_context *r300,
- const struct pipe_draw_info *info)
+ const struct pipe_draw_info *info,
+ int instance_id)
{
boolean alt_num_verts = r300->screen->caps.is_r500 &&
info->count > 65536;
@@ -710,7 +719,7 @@ static void r300_draw_arrays(struct r300_context *r300,
/* 9 spare dwords for emit_draw_arrays. Give up if the function fails. */
if (!r300_prepare_for_rendering(r300,
PREP_EMIT_STATES | PREP_VALIDATE_VBOS | PREP_EMIT_AOS,
- NULL, 9, start, 0))
+ NULL, 9, start, 0, instance_id))
return;
if (alt_num_verts || count <= 65535) {
@@ -727,13 +736,31 @@ static void r300_draw_arrays(struct r300_context *r300,
if (count) {
if (!r300_prepare_for_rendering(r300,
PREP_VALIDATE_VBOS | PREP_EMIT_AOS, NULL, 9,
- start, 0))
+ start, 0, instance_id))
return;
}
} while (count);
}
}
+static void r300_draw_arrays_instanced(struct r300_context *r300,
+ const struct pipe_draw_info *info)
+{
+ int i;
+
+ for (i = 0; i < info->instance_count; i++)
+ r300_draw_arrays(r300, info, i);
+}
+
+static void r300_draw_elements_instanced(struct r300_context *r300,
+ const struct pipe_draw_info *info)
+{
+ int i;
+
+ for (i = 0; i < info->instance_count; i++)
+ r300_draw_elements(r300, info, i);
+}
+
static void r300_draw_vbo(struct pipe_context* pipe,
const struct pipe_draw_info *dinfo)
{
@@ -767,20 +794,20 @@ static void r300_draw_vbo(struct pipe_context* pipe,
r300_resource(r300->index_buffer.buffer)->b.user_ptr) {
r300_draw_elements_immediate(r300, &info);
} else {
- r300_draw_elements(r300, &info);
+ r300_draw_elements(r300, &info, -1);
}
} else {
- assert(0);
+ r300_draw_elements_instanced(r300, &info);
}
} else {
if (info.instance_count <= 1) {
if (immd_is_good_idea(r300, info.count)) {
r300_draw_arrays_immediate(r300, &info);
} else {
- r300_draw_arrays(r300, &info);
+ r300_draw_arrays(r300, &info, -1);
}
} else {
- assert(0);
+ r300_draw_arrays_instanced(r300, &info);
}
}
@@ -998,12 +1025,12 @@ static void r300_render_draw_arrays(struct vbuf_render* render,
if (r300->draw_first_emitted) {
if (!r300_prepare_for_rendering(r300,
PREP_EMIT_STATES | PREP_EMIT_AOS_SWTCL,
- NULL, dwords, 0, 0))
+ NULL, dwords, 0, 0, -1))
return;
} else {
if (!r300_emit_states(r300,
PREP_EMIT_STATES | PREP_EMIT_AOS_SWTCL,
- NULL, 0, 0))
+ NULL, 0, 0, -1))
return;
}
@@ -1038,12 +1065,12 @@ static void r300_render_draw_elements(struct vbuf_render* render,
if (r300->draw_first_emitted) {
if (!r300_prepare_for_rendering(r300,
PREP_EMIT_STATES | PREP_EMIT_AOS_SWTCL | PREP_INDEXED,
- NULL, 256, 0, 0))
+ NULL, 256, 0, 0, -1))
return;
} else {
if (!r300_emit_states(r300,
PREP_EMIT_STATES | PREP_EMIT_AOS_SWTCL | PREP_INDEXED,
- NULL, 0, 0))
+ NULL, 0, 0, -1))
return;
}
@@ -1080,7 +1107,7 @@ static void r300_render_draw_elements(struct vbuf_render* render,
if (count) {
if (!r300_prepare_for_rendering(r300,
PREP_EMIT_AOS_SWTCL | PREP_INDEXED,
- NULL, 256, 0, 0))
+ NULL, 256, 0, 0, -1))
return;
end_cs_dwords = r300_get_num_cs_end_dwords(r300);
@@ -1184,7 +1211,7 @@ static void r300_blitter_draw_rectangle(struct blitter_context *blitter,
r300->clip_state.dirty = FALSE;
r300->viewport_state.dirty = FALSE;
- if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0))
+ if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1))
goto done;
DBG(r300, DBG_DRAW, "r300: draw_rectangle\n");
diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c
index 2f82f78..6c0dc99 100644
--- a/src/gallium/drivers/r300/r300_screen.c
+++ b/src/gallium/drivers/r300/r300_screen.c
@@ -112,6 +112,7 @@ static int r300_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
case PIPE_CAP_TEXTURE_MIRROR_CLAMP:
case PIPE_CAP_TEXTURE_MIRROR_REPEAT:
case PIPE_CAP_BLEND_EQUATION_SEPARATE:
+ case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
return 1;
case PIPE_CAP_TEXTURE_SWIZZLE:
return util_format_s3tc_enabled ? r300screen->caps.dxtc_swizzle : 1;
@@ -127,7 +128,6 @@ static int r300_get_param(struct pipe_screen* pscreen, enum pipe_cap param)
case PIPE_CAP_STREAM_OUTPUT:
case PIPE_CAP_PRIMITIVE_RESTART:
case PIPE_CAP_TGSI_INSTANCEID:
- case PIPE_CAP_VERTEX_ELEMENT_INSTANCE_DIVISOR:
case PIPE_CAP_ARRAY_TEXTURES:
return 0;
More information about the mesa-commit
mailing list