#include "util/u_memory.h"
#include "util/u_upload_mgr.h"
#include "util/u_prim.h"
+#include "util/u_draw_quad.h"
#include "r300_cs.h"
#include "r300_context.h"
#include <limits.h>
+#define IMMD_DWORDS 32
+
static uint32_t r300_translate_primitive(unsigned prim)
{
switch (prim) {
/* Emitted in flush. */
end_dwords += 26; /* emit_query_end */
+ end_dwords += r300->hyperz_state.size; /* emit_hyperz_end */
cs_dwords += end_dwords;
/* Reserve requested CS space. */
- if (!r300_check_cs(r300, cs_dwords)) {
+ if (cs_dwords > (r300->cs->ndw - r300->cs->cdw)) {
r300->context.flush(&r300->context, 0, NULL);
flushed = TRUE;
}
return FALSE;
}
- if (count > 10) {
+ if (count * r300->velems->vertex_size_dwords > IMMD_DWORDS) {
return FALSE;
}
/* We shouldn't map buffers referenced by CS, busy buffers,
* and ones placed in VRAM. */
- /* XXX Check for VRAM buffers. */
for (i = 0; i < vertex_element_count; i++) {
velem = &r300->velems->velem[i];
vbi = velem->vertex_buffer_index;
if (!checked[vbi]) {
vbuf = &r300->vertex_buffer[vbi];
+ if (!(r300_buffer(vbuf->buffer)->domain & R300_DOMAIN_GTT)) {
+ return FALSE;
+ }
+
if (r300_buffer_is_referenced(&r300->context,
vbuf->buffer,
R300_REF_CS | R300_REF_HW)) {
}
/*****************************************************************************
- * The emission of draw packets for r500. Older GPUs may use these functions *
- * after resolving fallback issues (e.g. stencil ref two-sided). *
+ * The HWTCL draw functions. *
****************************************************************************/
static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
struct pipe_vertex_element* velem;
struct pipe_vertex_buffer* vbuf;
unsigned vertex_element_count = r300->velems->count;
- unsigned i, v, vbi, dw, elem_offset, dwords;
+ unsigned i, v, vbi, dwords;
/* Size of the vertex, in dwords. */
- unsigned vertex_size = 0;
-
- /* Offsets of the attribute, in dwords, from the start of the vertex. */
- unsigned offset[PIPE_MAX_ATTRIBS];
+ unsigned vertex_size = r300->velems->vertex_size_dwords;
/* Size of the vertex element, in dwords. */
unsigned size[PIPE_MAX_ATTRIBS];
/* Stride to the same attrib in the next vertex in the vertex buffer,
* in dwords. */
- unsigned stride[PIPE_MAX_ATTRIBS] = {0};
+ unsigned stride[PIPE_MAX_ATTRIBS];
/* Mapped vertex buffers. */
- uint32_t* map[PIPE_MAX_ATTRIBS] = {0};
- struct pipe_transfer* transfer[PIPE_MAX_ATTRIBS] = {NULL};
+ uint32_t* map[PIPE_MAX_ATTRIBS];
+ uint32_t* mapelem[PIPE_MAX_ATTRIBS];
+ struct pipe_transfer* transfer[PIPE_MAX_ATTRIBS] = {0};
CS_LOCALS(r300);
/* Calculate the vertex size, offsets, strides etc. and map the buffers. */
for (i = 0; i < vertex_element_count; i++) {
velem = &r300->velems->velem[i];
- offset[i] = velem->src_offset / 4;
size[i] = r300->velems->hw_format_size[i] / 4;
- vertex_size += size[i];
vbi = velem->vertex_buffer_index;
+ vbuf = &r300->vertex_buffer[vbi];
+ stride[i] = vbuf->stride / 4;
/* Map the buffer. */
- if (!map[vbi]) {
- vbuf = &r300->vertex_buffer[vbi];
+ if (!transfer[vbi]) {
map[vbi] = (uint32_t*)pipe_buffer_map(&r300->context,
vbuf->buffer,
PIPE_TRANSFER_READ,
&transfer[vbi]);
- map[vbi] += vbuf->buffer_offset / 4;
- stride[vbi] = vbuf->stride / 4;
+ map[vbi] += (vbuf->buffer_offset / 4) + stride[i] * start;
}
+ mapelem[i] = map[vbi] + (velem->src_offset / 4);
}
dwords = 9 + count * vertex_size;
/* Emit vertices. */
for (v = 0; v < count; v++) {
for (i = 0; i < vertex_element_count; i++) {
- velem = &r300->velems->velem[i];
- vbi = velem->vertex_buffer_index;
- elem_offset = offset[i] + stride[vbi] * (v + start);
-
- for (dw = 0; dw < size[i]; dw++) {
- OUT_CS(map[vbi][elem_offset + dw]);
- }
+ OUT_CS_TABLE(&mapelem[i][stride[i] * v], size[i]);
}
}
END_CS;
for (i = 0; i < vertex_element_count; i++) {
vbi = r300->velems->velem[i].vertex_buffer_index;
- if (map[vbi]) {
+ if (transfer[vbi]) {
vbuf = &r300->vertex_buffer[vbi];
pipe_buffer_unmap(&r300->context, vbuf->buffer, transfer[vbi]);
- map[vbi] = NULL;
+ transfer[vbi] = NULL;
}
}
}
(0 << R300_INDX_BUFFER_SKIP_SHIFT));
OUT_CS(offset_dwords << 2);
OUT_CS_BUF_RELOC(indexBuffer, count_dwords,
- r300_buffer(indexBuffer)->domain, 0, 0);
+ r300_buffer(indexBuffer)->domain, 0);
END_CS;
}
unsigned short_count;
int buffer_offset = 0, index_offset = 0; /* for index bias emulation */
boolean translate = FALSE;
+ unsigned new_offset;
if (r300->skip_rendering) {
return;
return;
}
+ /* Index buffer range checking. */
+ if ((start + count) * indexSize > indexBuffer->width0) {
+ fprintf(stderr, "r300: Invalid index buffer range. Skipping rendering.\n");
+ return;
+ }
+
/* Set up fallback for incompatible vertex layout if needed. */
if (r300->incompatible_vb_layout || r300->velems->incompatible_layout) {
r300_begin_vertex_translate(r300);
&start, count);
r300_update_derived_state(r300);
- r300_upload_index_buffer(r300, &indexBuffer, indexSize, start, count);
+ r300_upload_index_buffer(r300, &indexBuffer, indexSize, start, count, &new_offset);
+ start = new_offset;
/* 15 dwords for emit_draw_elements */
r300_prepare_for_rendering(r300,
PREP_FIRST_DRAW | PREP_VALIDATE_VBOS | PREP_EMIT_AOS | PREP_INDEXED,
indexBuffer, 15, buffer_offset, indexBias, NULL);
- u_upload_flush(r300->upload_vb);
- u_upload_flush(r300->upload_ib);
if (alt_num_verts || count <= 65535) {
r300_emit_draw_elements(r300, indexBuffer, indexSize,
- minIndex, maxIndex, mode, start, count);
+ minIndex, maxIndex, mode, start, count);
} else {
do {
short_count = MIN2(count, 65534);
}
}
+static void r300_draw_vbo(struct pipe_context* pipe,
+ const struct pipe_draw_info *info)
+{
+ struct r300_context* r300 = r300_context(pipe);
+
+ if (info->indexed && r300->index_buffer.buffer) {
+ unsigned offset;
+
+ assert(r300->index_buffer.offset % r300->index_buffer.index_size == 0);
+ offset = r300->index_buffer.offset / r300->index_buffer.index_size;
+
+ r300_draw_range_elements(pipe,
+ r300->index_buffer.buffer,
+ r300->index_buffer.index_size,
+ info->index_bias,
+ info->min_index,
+ info->max_index,
+ info->mode,
+ info->start + offset,
+ info->count);
+ }
+ else {
+ r300_draw_arrays(pipe,
+ info->mode,
+ info->start,
+ info->count);
+ }
+}
+
/****************************************************************************
* The rest of this file is for SW TCL rendering only. Please be polite and *
* keep these functions separated so that they are easier to locate. ~C. *
***************************************************************************/
-/* SW TCL arrays, using Draw. */
-static void r300_swtcl_draw_arrays(struct pipe_context* pipe,
- unsigned mode,
- unsigned start,
- unsigned count)
+/* SW TCL elements, using Draw. */
+static void r300_swtcl_draw_vbo(struct pipe_context* pipe,
+ const struct pipe_draw_info *info)
{
struct r300_context* r300 = r300_context(pipe);
struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS];
+ struct pipe_transfer *ib_transfer;
+ unsigned count = info->count;
int i;
+ void* indices = NULL;
if (r300->skip_rendering) {
return;
}
- if (!u_trim_pipe_prim(mode, &count)) {
+ if (!u_trim_pipe_prim(info->mode, &count)) {
return;
}
void* buf = pipe_buffer_map(pipe,
r300->vertex_buffer[i].buffer,
PIPE_TRANSFER_READ,
- &vb_transfer[i]);
+ &vb_transfer[i]);
draw_set_mapped_vertex_buffer(r300->draw, i, buf);
}
- draw_set_mapped_element_buffer(r300->draw, 0, 0, NULL);
+ if (info->indexed && r300->index_buffer.buffer) {
+ indices = pipe_buffer_map(pipe, r300->index_buffer.buffer,
+ PIPE_TRANSFER_READ, &ib_transfer);
+ if (indices)
+ indices += r300->index_buffer.offset;
+ }
+
+ draw_set_mapped_element_buffer_range(r300->draw, (indices) ?
+ r300->index_buffer.index_size : 0,
+ info->index_bias,
+ info->min_index,
+ info->max_index,
+ indices);
- draw_arrays(r300->draw, mode, start, count);
+ draw_arrays(r300->draw, info->mode, info->start, count);
/* XXX Not sure whether this is the best fix.
* It prevents CS from being rejected and weird assertion failures. */
for (i = 0; i < r300->vertex_buffer_count; i++) {
pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer,
- vb_transfer[i]);
+ vb_transfer[i]);
draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
}
+
+ if (ib_transfer) {
+ pipe_buffer_unmap(pipe, r300->index_buffer.buffer, ib_transfer);
+ draw_set_mapped_element_buffer_range(r300->draw, 0, 0, info->start,
+ info->start + count - 1, NULL);
+ }
}
/* SW TCL elements, using Draw. */
unsigned count)
{
struct r300_context* r300 = r300_context(pipe);
- struct pipe_transfer *vb_transfer[PIPE_MAX_ATTRIBS];
- struct pipe_transfer *ib_transfer;
- int i;
- void* indices;
-
- if (r300->skip_rendering) {
- return;
- }
-
- if (!u_trim_pipe_prim(mode, &count)) {
- return;
- }
-
- r300_update_derived_state(r300);
-
- for (i = 0; i < r300->vertex_buffer_count; i++) {
- void* buf = pipe_buffer_map(pipe,
- r300->vertex_buffer[i].buffer,
- PIPE_TRANSFER_READ,
- &vb_transfer[i]);
- draw_set_mapped_vertex_buffer(r300->draw, i, buf);
+ struct pipe_draw_info info;
+ struct pipe_index_buffer saved_ib, ib;
+
+ util_draw_init_info(&info);
+ info.mode = mode;
+ info.start = start;
+ info.count = count;
+ info.index_bias = indexBias;
+ info.min_index = minIndex;
+ info.max_index = maxIndex;
+
+ if (indexBuffer) {
+ info.indexed = TRUE;
+
+ saved_ib = r300->index_buffer;
+ ib.buffer = indexBuffer;
+ ib.offset = 0;
+ ib.index_size = indexSize;
+ pipe->set_index_buffer(pipe, &ib);
}
- indices = pipe_buffer_map(pipe, indexBuffer,
- PIPE_TRANSFER_READ, &ib_transfer);
- draw_set_mapped_element_buffer_range(r300->draw, indexSize, indexBias,
- minIndex, maxIndex, indices);
-
- draw_arrays(r300->draw, mode, start, count);
+ r300_swtcl_draw_vbo(pipe, &info);
- /* XXX Not sure whether this is the best fix.
- * It prevents CS from being rejected and weird assertion failures. */
- draw_flush(r300->draw);
-
- for (i = 0; i < r300->vertex_buffer_count; i++) {
- pipe_buffer_unmap(pipe, r300->vertex_buffer[i].buffer,
- vb_transfer[i]);
- draw_set_mapped_vertex_buffer(r300->draw, i, NULL);
- }
+ if (indexBuffer)
+ pipe->set_index_buffer(pipe, &saved_ib);
+}
- pipe_buffer_unmap(pipe, indexBuffer,
- ib_transfer);
- draw_set_mapped_element_buffer_range(r300->draw, 0, 0,
- start, start + count - 1,
- NULL);
+static void r300_swtcl_draw_arrays(struct pipe_context* pipe,
+ unsigned mode,
+ unsigned start,
+ unsigned count)
+{
+ r300_swtcl_draw_range_elements(pipe, NULL, 0, 0,
+ start, start + count -1, mode, start, count);
}
/* Object for rendering using Draw. */
unsigned dwords = 6;
CS_LOCALS(r300);
-
(void) i; (void) ptr;
r300_prepare_for_rendering(r300, PREP_FIRST_DRAW | PREP_EMIT_AOS_SWTCL,
NULL, dwords, 0, 0, NULL);
- DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
+ DBG(r300, DBG_DRAW, "r300: render_draw_arrays (count: %d)\n", count);
/* Uncomment to dump all VBOs rendered through this interface.
* Slow and noisy!
unsigned max_index = (r300render->vbo_size - r300render->vbo_offset) /
(r300render->r300->vertex_info.size * 4) - 1;
unsigned short_count;
- struct r300_cs_info cs_info;
+ unsigned free_dwords;
CS_LOCALS(r300);
+ DBG(r300, DBG_DRAW, "r300: render_draw_elements (count: %d)\n", count);
/* Reserve at least 256 dwords.
*
NULL, 256, 0, 0, &end_cs_dwords);
while (count) {
- r300->rws->get_cs_info(r300->rws, &cs_info);
+ free_dwords = r300->cs->ndw - r300->cs->cdw;
- short_count = MIN2(count, (cs_info.free - end_cs_dwords - 6) * 2);
+ short_count = MIN2(count, (free_dwords - end_cs_dwords - 6) * 2);
BEGIN_CS(6 + (short_count+1)/2);
OUT_CS_REG(R300_GA_COLOR_CONTROL,
return stage;
}
+/****************************************************************************
+ * End of SW TCL functions *
+ ***************************************************************************/
+
+/* If we used a quad to draw a rectangle, the pixels on the main diagonal
+ * would be computed and stored twice, which makes the clear/copy codepaths
+ * somewhat inefficient. Instead we use a rectangular point sprite. */
+static void r300_blitter_draw_rectangle(struct blitter_context *blitter,
+ unsigned x1, unsigned y1,
+ unsigned x2, unsigned y2,
+ float depth,
+ enum blitter_attrib_type type,
+ const float attrib[4])
+{
+ struct r300_context *r300 = r300_context(util_blitter_get_pipe(blitter));
+ unsigned last_sprite_coord_enable = r300->sprite_coord_enable;
+ unsigned width = x2 - x1;
+ unsigned height = y2 - y1;
+ unsigned vertex_size =
+ type == UTIL_BLITTER_ATTRIB_COLOR || !r300->draw ? 8 : 4;
+ unsigned dwords = 13 + vertex_size +
+ (type == UTIL_BLITTER_ATTRIB_TEXCOORD ? 7 : 0);
+ const float zeros[4] = {0, 0, 0, 0};
+ CS_LOCALS(r300);
+
+ if (type == UTIL_BLITTER_ATTRIB_TEXCOORD)
+ r300->sprite_coord_enable = 1;
+
+ r300_update_derived_state(r300);
+
+ /* Mark some states we don't care about as non-dirty. */
+ r300->clip_state.dirty = FALSE;
+ r300->viewport_state.dirty = FALSE;
+
+ r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0, NULL);
+
+ DBG(r300, DBG_DRAW, "r300: draw_rectangle\n");
+
+ BEGIN_CS(dwords);
+ /* Set up GA. */
+ OUT_CS_REG(R300_GA_POINT_SIZE, (height * 6) | ((width * 6) << 16));
+
+ if (type == UTIL_BLITTER_ATTRIB_TEXCOORD) {
+ /* Set up the GA to generate texcoords. */
+ OUT_CS_REG(R300_GB_ENABLE, R300_GB_POINT_STUFF_ENABLE |
+ (R300_GB_TEX_STR << R300_GB_TEX0_SOURCE_SHIFT));
+ OUT_CS_REG_SEQ(R300_GA_POINT_S0, 4);
+ OUT_CS_32F(attrib[0]);
+ OUT_CS_32F(attrib[3]);
+ OUT_CS_32F(attrib[2]);
+ OUT_CS_32F(attrib[1]);
+ }
+
+ /* Set up VAP controls. */
+ OUT_CS_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
+ OUT_CS_REG(R300_VAP_VTE_CNTL, R300_VTX_XY_FMT | R300_VTX_Z_FMT);
+ OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
+ OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
+ OUT_CS(1);
+ OUT_CS(0);
+
+ /* Draw. */
+ OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, vertex_size);
+ OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (1 << 16) |
+ R300_VAP_VF_CNTL__PRIM_POINTS);
+
+ OUT_CS_32F(x1 + width * 0.5f);
+ OUT_CS_32F(y1 + height * 0.5f);
+ OUT_CS_32F(depth);
+ OUT_CS_32F(1);
+
+ if (vertex_size == 8) {
+ if (!attrib)
+ attrib = zeros;
+ OUT_CS_TABLE(attrib, 4);
+ }
+ END_CS;
+
+ /* Restore the state. */
+ r300->clip_state.dirty = TRUE;
+ r300->rs_state.dirty = TRUE;
+ r300->viewport_state.dirty = TRUE;
+
+ r300->sprite_coord_enable = last_sprite_coord_enable;
+}
+
+static void r300_resource_resolve(struct pipe_context* pipe,
+ struct pipe_resource* dest,
+ struct pipe_subresource subdest,
+ struct pipe_resource* src,
+ struct pipe_subresource subsrc)
+{
+ struct r300_context* r300 = r300_context(pipe);
+ struct r300_aa_state *aa = (struct r300_aa_state*)r300->aa_state.state;
+ struct pipe_surface* srcsurf = src->screen->get_tex_surface(src->screen,
+ src, subsrc.face, subsrc.level, 0, 0);
+ float color[] = {0, 0, 0, 0};
+
+ DBG(r300, DBG_DRAW, "r300: Resolving resource...\n");
+
+ /* Enable AA resolve. */
+ aa->dest = r300_surface(
+ dest->screen->get_tex_surface(dest->screen, dest, subdest.face,
+ subdest.level, 0, 0));
+
+ aa->aaresolve_ctl =
+ R300_RB3D_AARESOLVE_CTL_AARESOLVE_MODE_RESOLVE |
+ R300_RB3D_AARESOLVE_CTL_AARESOLVE_ALPHA_AVERAGE;
+ r300->aa_state.size = 12;
+ r300->aa_state.dirty = TRUE;
+
+ /* Resolve the surface. */
+ r300->context.clear_render_target(pipe,
+ srcsurf, color, 0, 0, src->width0, src->height0);
+
+ /* Disable AA resolve. */
+ aa->aaresolve_ctl = 0;
+ r300->aa_state.size = 4;
+ r300->aa_state.dirty = TRUE;
+
+ pipe_surface_reference((struct pipe_surface**)&srcsurf, NULL);
+ pipe_surface_reference((struct pipe_surface**)&aa->dest, NULL);
+}
+
void r300_init_render_functions(struct r300_context *r300)
{
/* Set generic functions. */
if (r300->screen->caps.has_tcl) {
r300->context.draw_arrays = r300_draw_arrays;
r300->context.draw_range_elements = r300_draw_range_elements;
+ r300->context.draw_vbo = r300_draw_vbo;
} else {
r300->context.draw_arrays = r300_swtcl_draw_arrays;
r300->context.draw_range_elements = r300_swtcl_draw_range_elements;
+ r300->context.draw_vbo = r300_swtcl_draw_vbo;
}
+ r300->context.resource_resolve = r300_resource_resolve;
+ r300->blitter->draw_rectangle = r300_blitter_draw_rectangle;
+
/* Plug in the two-sided stencil reference value fallback if needed. */
if (!r300->screen->caps.is_r500)
r300_plug_in_stencil_ref_fallback(r300);