void r300_emit_vertex_stream_state(struct r300_context* r300,
unsigned size, void* state)
{
struct r300_vertex_stream_state *streams =
(struct r300_vertex_stream_state*)state;
unsigned i;
CS_LOCALS(r300);
if (DBG_ON(r300, DBG_PSC)) {
fprintf(stderr, "r300: PSC emit:\n");
for (i = 0; i < streams->count; i++) {
fprintf(stderr, " : prog_stream_cntl%d: 0x%08x\n", i,
streams->vap_prog_stream_cntl[i]);
}
for (i = 0; i < streams->count; i++) {
fprintf(stderr, " : prog_stream_cntl_ext%d: 0x%08x\n", i,
streams->vap_prog_stream_cntl_ext[i]);
}
}
BEGIN_CS(size);
OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_0, streams->count);
OUT_CS_TABLE(streams->vap_prog_stream_cntl, streams->count);
OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_EXT_0, streams->count);
OUT_CS_TABLE(streams->vap_prog_stream_cntl_ext, streams->count);
END_CS;
}
void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state)
{
struct r300_vertex_shader* vs = (struct r300_vertex_shader*)state;
struct r300_vertex_program_code* code = &vs->code;
struct r300_screen* r300screen = r300->screen;
unsigned instruction_count = code->length / 4;
unsigned vtx_mem_size = r300screen->caps.is_r500 ? 128 : 72;
unsigned input_count = MAX2(util_bitcount(code->InputsRead), 1);
unsigned output_count = MAX2(util_bitcount(code->OutputsWritten), 1);
unsigned temp_count = MAX2(code->num_temporaries, 1);
unsigned pvs_num_slots = MIN3(vtx_mem_size / input_count,
vtx_mem_size / output_count, 10);
unsigned pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 5);
CS_LOCALS(r300);
BEGIN_CS(size);
/* R300_VAP_PVS_CODE_CNTL_0
* R300_VAP_PVS_CONST_CNTL
* R300_VAP_PVS_CODE_CNTL_1
* See the r5xx docs for instructions on how to use these. */
OUT_CS_REG(R300_VAP_PVS_CODE_CNTL_0, R300_PVS_FIRST_INST(0) |
R300_PVS_XYZW_VALID_INST(instruction_count - 1) |
R300_PVS_LAST_INST(instruction_count - 1));
OUT_CS_REG(R300_VAP_PVS_CODE_CNTL_1, instruction_count - 1);
OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG, 0);
OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, code->length);
OUT_CS_TABLE(code->body.d, code->length);
OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(pvs_num_slots) |
R300_PVS_NUM_CNTLRS(pvs_num_controllers) |
R300_PVS_NUM_FPUS(r300screen->caps.num_vert_fpus) |
R300_PVS_VF_MAX_VTX_NUM(12) |
(r300->clip_halfz ? R300_DX_CLIP_SPACE_DEF : 0) |
(r300screen->caps.is_r500 ? R500_TCL_STATE_OPTIMIZATION : 0));
/* Emit flow control instructions. Even if there are no fc instructions,
* we still need to write the registers to make sure they are cleared. */
OUT_CS_REG(R300_VAP_PVS_FLOW_CNTL_OPC, code->fc_ops);
if (r300screen->caps.is_r500) {
OUT_CS_REG_SEQ(R500_VAP_PVS_FLOW_CNTL_ADDRS_LW_0, R300_VS_MAX_FC_OPS * 2);
OUT_CS_TABLE(code->fc_op_addrs.r500, R300_VS_MAX_FC_OPS * 2);
} else {
OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_ADDRS_0, R300_VS_MAX_FC_OPS);
OUT_CS_TABLE(code->fc_op_addrs.r300, R300_VS_MAX_FC_OPS);
}
OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_LOOP_INDEX_0, R300_VS_MAX_FC_OPS);
OUT_CS_TABLE(code->fc_loop_index, R300_VS_MAX_FC_OPS);
END_CS;
}
void r300_emit_rs_block_state(struct r300_context* r300,
unsigned size, void* state)
{
struct r300_rs_block* rs = (struct r300_rs_block*)state;
unsigned i;
/* It's the same for both INST and IP tables */
unsigned count = (rs->inst_count & R300_RS_INST_COUNT_MASK) + 1;
CS_LOCALS(r300);
if (DBG_ON(r300, DBG_RS_BLOCK)) {
r500_dump_rs_block(rs);
fprintf(stderr, "r300: RS emit:\n");
for (i = 0; i < count; i++)
fprintf(stderr, " : ip %d: 0x%08x\n", i, rs->ip[i]);
for (i = 0; i < count; i++)
fprintf(stderr, " : inst %d: 0x%08x\n", i, rs->inst[i]);
fprintf(stderr, " : count: 0x%08x inst_count: 0x%08x\n",
rs->count, rs->inst_count);
}
BEGIN_CS(size);
OUT_CS_REG_SEQ(R300_VAP_VTX_STATE_CNTL, 2);
OUT_CS(rs->vap_vtx_state_cntl);
OUT_CS(rs->vap_vsm_vtx_assm);
OUT_CS_REG_SEQ(R300_VAP_OUTPUT_VTX_FMT_0, 2);
OUT_CS(rs->vap_out_vtx_fmt[0]);
OUT_CS(rs->vap_out_vtx_fmt[1]);
OUT_CS_REG_SEQ(R300_GB_ENABLE, 1);
OUT_CS(rs->gb_enable);
if (r300->screen->caps.is_r500) {
OUT_CS_REG_SEQ(R500_RS_IP_0, count);
} else {
OUT_CS_REG_SEQ(R300_RS_IP_0, count);
}
OUT_CS_TABLE(rs->ip, count);
OUT_CS_REG_SEQ(R300_RS_COUNT, 2);
OUT_CS(rs->count);
OUT_CS(rs->inst_count);
if (r300->screen->caps.is_r500) {
OUT_CS_REG_SEQ(R500_RS_INST_0, count);
} else {
OUT_CS_REG_SEQ(R300_RS_INST_0, count);
}
OUT_CS_TABLE(rs->inst, count);
END_CS;
}
void r300_emit_dsa_state(struct r300_context* r300, unsigned size, void* state)
{
struct r300_dsa_state* dsa = (struct r300_dsa_state*)state;
struct pipe_framebuffer_state* fb =
(struct pipe_framebuffer_state*)r300->fb_state.state;
boolean is_r500 = r300->screen->caps.is_r500;
CS_LOCALS(r300);
uint32_t alpha_func = dsa->alpha_function;
/* Choose the alpha ref value between 8-bit (FG_ALPHA_FUNC.AM_VAL) and
* 16-bit (FG_ALPHA_VALUE). */
if (is_r500 && (alpha_func & R300_FG_ALPHA_FUNC_ENABLE)) {
struct pipe_surface *cb = fb->nr_cbufs ? r300_get_nonnull_cb(fb, 0) : NULL;
if (cb &&
(cb->format == PIPE_FORMAT_R16G16B16A16_FLOAT ||
cb->format == PIPE_FORMAT_R16G16B16X16_FLOAT)) {
alpha_func |= R500_FG_ALPHA_FUNC_FP16_ENABLE;
} else {
alpha_func |= R500_FG_ALPHA_FUNC_8BIT;
}
}
/* Setup alpha-to-coverage. */
if (r300->alpha_to_coverage && r300->msaa_enable) {
/* Always set 3/6, it improves precision even for 2x and 4x MSAA. */
alpha_func |= R300_FG_ALPHA_FUNC_MASK_ENABLE |
R300_FG_ALPHA_FUNC_CFG_3_OF_6;
}
BEGIN_CS(size);
OUT_CS_REG(R300_FG_ALPHA_FUNC, alpha_func);
OUT_CS_TABLE(fb->zsbuf ? &dsa->cb_begin : dsa->cb_zb_no_readwrite, size-2);
END_CS;
}
void r300_emit_rs_state(struct r300_context* r300, unsigned size, void* state)
{
struct r300_rs_state* rs = state;
CS_LOCALS(r300);
BEGIN_CS(size);
OUT_CS_TABLE(rs->cb_main, RS_STATE_MAIN_SIZE);
if (rs->polygon_offset_enable) {
if (r300->zbuffer_bpp == 16) {
OUT_CS_TABLE(rs->cb_poly_offset_zb16, 5);
} else {
OUT_CS_TABLE(rs->cb_poly_offset_zb24, 5);
}
}
END_CS;
}
void r500_emit_fs_rc_constant_state(struct r300_context* r300, unsigned size, void *state)
{
struct r300_fragment_shader *fs = r300_fs(r300);
struct rc_constant_list *constants = &fs->shader->code.constants;
unsigned i;
unsigned count = fs->shader->rc_state_count;
unsigned first = fs->shader->externals_count;
unsigned end = constants->Count;
CS_LOCALS(r300);
if (count == 0)
return;
BEGIN_CS(size);
for(i = first; i < end; ++i) {
if (constants->Constants[i].Type == RC_CONSTANT_STATE) {
float data[4];
get_rc_constant_state(data, r300, &constants->Constants[i]);
OUT_CS_REG(R500_GA_US_VECTOR_INDEX,
R500_GA_US_VECTOR_INDEX_TYPE_CONST |
(i & R500_GA_US_VECTOR_INDEX_MASK));
OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, 4);
OUT_CS_TABLE(data, 4);
}
}
END_CS;
}
void r300_emit_vs_constants(struct r300_context* r300,
unsigned size, void *state)
{
unsigned count =
((struct r300_vertex_shader*)r300->vs_state.state)->externals_count;
struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state;
struct r300_vertex_shader *vs = (struct r300_vertex_shader*)r300->vs_state.state;
unsigned i;
int imm_first = vs->externals_count;
int imm_end = vs->code.constants.Count;
int imm_count = vs->immediates_count;
CS_LOCALS(r300);
BEGIN_CS(size);
OUT_CS_REG(R300_VAP_PVS_CONST_CNTL,
R300_PVS_CONST_BASE_OFFSET(buf->buffer_base) |
R300_PVS_MAX_CONST_ADDR(MAX2(imm_end - 1, 0)));
if (vs->externals_count) {
OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG,
(r300->screen->caps.is_r500 ?
R500_PVS_CONST_START : R300_PVS_CONST_START) + buf->buffer_base);
OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, count * 4);
if (buf->remap_table){
for (i = 0; i < count; i++) {
uint32_t *data = &buf->ptr[buf->remap_table[i]*4];
OUT_CS_TABLE(data, 4);
}
} else {
OUT_CS_TABLE(buf->ptr, count * 4);
}
}
/* Emit immediates. */
if (imm_count) {
OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG,
(r300->screen->caps.is_r500 ?
R500_PVS_CONST_START : R300_PVS_CONST_START) +
buf->buffer_base + imm_first);
OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, imm_count * 4);
for (i = imm_first; i < imm_end; i++) {
const float *data = vs->code.constants.Constants[i].u.Immediate;
OUT_CS_TABLE(data, 4);
}
}
END_CS;
}
void r300_emit_viewport_state(struct r300_context* r300,
unsigned size, void* state)
{
struct r300_viewport_state* viewport = (struct r300_viewport_state*)state;
CS_LOCALS(r300);
BEGIN_CS(size);
OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6);
OUT_CS_TABLE(&viewport->xscale, 6);
OUT_CS_REG(R300_VAP_VTE_CNTL, viewport->vte_control);
END_CS;
}
void r500_emit_fs_constants(struct r300_context* r300, unsigned size, void *state)
{
struct r300_fragment_shader *fs = r300_fs(r300);
struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state;
unsigned count = fs->shader->externals_count;
CS_LOCALS(r300);
if (count == 0)
return;
BEGIN_CS(size);
OUT_CS_REG(R500_GA_US_VECTOR_INDEX, R500_GA_US_VECTOR_INDEX_TYPE_CONST);
OUT_CS_ONE_REG(R500_GA_US_VECTOR_DATA, count * 4);
if (buf->remap_table){
for (unsigned i = 0; i < count; i++) {
uint32_t *data = &buf->ptr[buf->remap_table[i]*4];
OUT_CS_TABLE(data, 4);
}
} else {
OUT_CS_TABLE(buf->ptr, count * 4);
}
END_CS;
}
void r300_emit_gpu_flush(struct r300_context *r300, unsigned size, void *state)
{
struct r300_gpu_flush *gpuflush = (struct r300_gpu_flush*)state;
struct pipe_framebuffer_state* fb =
(struct pipe_framebuffer_state*)r300->fb_state.state;
uint32_t height = fb->height;
uint32_t width = fb->width;
CS_LOCALS(r300);
if (r300->cbzb_clear) {
struct r300_surface *surf = r300_surface(fb->cbufs[0]);
height = surf->cbzb_height;
width = surf->cbzb_width;
}
DBG(r300, DBG_SCISSOR,
"r300: Scissor width: %i, height: %i, CBZB clear: %s\n",
width, height, r300->cbzb_clear ? "YES" : "NO");
BEGIN_CS(size);
/* Set up scissors.
* By writing to the SC registers, SC & US assert idle. */
OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL, 2);
if (r300->screen->caps.is_r500) {
OUT_CS(0);
OUT_CS(((width - 1) << R300_SCISSORS_X_SHIFT) |
((height - 1) << R300_SCISSORS_Y_SHIFT));
} else {
OUT_CS((1440 << R300_SCISSORS_X_SHIFT) |
(1440 << R300_SCISSORS_Y_SHIFT));
OUT_CS(((width + 1440-1) << R300_SCISSORS_X_SHIFT) |
((height + 1440-1) << R300_SCISSORS_Y_SHIFT));
}
/* Flush CB & ZB caches and wait until the 3D engine is idle and clean. */
OUT_CS_TABLE(gpuflush->cb_flush_clean, 6);
END_CS;
}
static void r300_draw_elements_immediate(struct r300_context *r300,
const struct pipe_draw_info *info)
{
const uint8_t *ptr1;
const uint16_t *ptr2;
const uint32_t *ptr4;
unsigned index_size = r300->index_buffer.index_size;
unsigned i, count_dwords = index_size == 4 ? info->count :
(info->count + 1) / 2;
CS_LOCALS(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_VARRAYS |
PREP_INDEXED, NULL, 2+count_dwords, 0, info->index_bias, -1))
return;
r300_emit_draw_init(r300, info->mode, info->max_index);
BEGIN_CS(2 + count_dwords);
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, count_dwords);
switch (index_size) {
case 1:
ptr1 = (uint8_t*)r300->index_buffer.user_buffer;
ptr1 += info->start;
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) |
r300_translate_primitive(info->mode));
if (info->index_bias && !r300->screen->caps.is_r500) {
for (i = 0; i < info->count-1; i += 2)
OUT_CS(((ptr1[i+1] + info->index_bias) << 16) |
(ptr1[i] + info->index_bias));
if (info->count & 1)
OUT_CS(ptr1[i] + info->index_bias);
} else {
for (i = 0; i < info->count-1; i += 2)
OUT_CS(((ptr1[i+1]) << 16) |
(ptr1[i] ));
if (info->count & 1)
OUT_CS(ptr1[i]);
}
break;
case 2:
ptr2 = (uint16_t*)r300->index_buffer.user_buffer;
ptr2 += info->start;
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) |
r300_translate_primitive(info->mode));
if (info->index_bias && !r300->screen->caps.is_r500) {
for (i = 0; i < info->count-1; i += 2)
OUT_CS(((ptr2[i+1] + info->index_bias) << 16) |
(ptr2[i] + info->index_bias));
if (info->count & 1)
OUT_CS(ptr2[i] + info->index_bias);
} else {
OUT_CS_TABLE(ptr2, count_dwords);
}
break;
case 4:
ptr4 = (uint32_t*)r300->index_buffer.user_buffer;
ptr4 += info->start;
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (info->count << 16) |
R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
r300_translate_primitive(info->mode));
if (info->index_bias && !r300->screen->caps.is_r500) {
for (i = 0; i < info->count; i++)
OUT_CS(ptr4[i] + info->index_bias);
} else {
OUT_CS_TABLE(ptr4, count_dwords);
}
break;
}
END_CS;
}
static void r300_draw_arrays_immediate(struct r300_context *r300,
const struct pipe_draw_info *info)
{
struct pipe_vertex_element* velem;
struct pipe_vertex_buffer* vbuf;
unsigned vertex_element_count = r300->velems->count;
unsigned i, v, vbi;
/* Size of the vertex, in dwords. */
unsigned vertex_size = r300->velems->vertex_size_dwords;
/* The number of dwords for this draw operation. */
unsigned dwords = 4 + info->count * vertex_size;
/* 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];
/* Mapped vertex buffers. */
uint32_t* map[PIPE_MAX_ATTRIBS] = {0};
uint32_t* mapelem[PIPE_MAX_ATTRIBS];
CS_LOCALS(r300);
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. */
for (i = 0; i < vertex_element_count; i++) {
velem = &r300->velems->velem[i];
size[i] = r300->velems->format_size[i] / 4;
vbi = velem->vertex_buffer_index;
vbuf = &r300->vertex_buffer[vbi];
stride[i] = vbuf->stride / 4;
/* Map the buffer. */
if (!map[vbi]) {
map[vbi] = (uint32_t*)r300->rws->buffer_map(
r300_resource(vbuf->buffer)->cs_buf,
r300->cs, PIPE_TRANSFER_READ | PIPE_TRANSFER_UNSYNCHRONIZED);
map[vbi] += (vbuf->buffer_offset / 4) + stride[i] * info->start;
}
mapelem[i] = map[vbi] + (velem->src_offset / 4);
}
r300_emit_draw_init(r300, info->mode, info->count-1);
BEGIN_CS(dwords);
OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, info->count * vertex_size);
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (info->count << 16) |
r300_translate_primitive(info->mode));
/* Emit vertices. */
for (v = 0; v < info->count; v++) {
for (i = 0; i < vertex_element_count; i++) {
OUT_CS_TABLE(&mapelem[i][stride[i] * v], size[i]);
}
}
END_CS;
}
/* This functions is used to draw a rectangle for the blitter module.
*
* If we rendered a quad, 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. */
void r300_blitter_draw_rectangle(struct blitter_context *blitter,
int x1, int y1, int x2, int y2,
float depth,
enum blitter_attrib_type type,
const union pipe_color_union *attrib)
{
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);
static const union pipe_color_union zeros;
CS_LOCALS(r300);
/* XXX workaround for a lockup in MSAA resolve on SWTCL chipsets, this
* function most probably doesn't handle type=NONE correctly */
if (!r300->screen->caps.has_tcl && type == UTIL_BLITTER_ATTRIB_NONE) {
util_blitter_draw_rectangle(blitter, x1, y1, x2, y2, depth, type, attrib);
return;
}
if (r300->skip_rendering)
return;
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->viewport_state.dirty = FALSE;
if (!r300_prepare_for_rendering(r300, PREP_EMIT_STATES, NULL, dwords, 0, 0, -1))
goto done;
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->f[0]);
OUT_CS_32F(attrib->f[3]);
OUT_CS_32F(attrib->f[2]);
OUT_CS_32F(attrib->f[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->f, 4);
}
END_CS;
done:
/* Restore the state. */
r300_mark_atom_dirty(r300, &r300->rs_state);
r300_mark_atom_dirty(r300, &r300->viewport_state);
r300->sprite_coord_enable = last_sprite_coord_enable;
}
static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
unsigned mode,
unsigned start,
unsigned count)
{
struct pipe_vertex_element* velem;
struct pipe_vertex_buffer* vbuf;
unsigned vertex_element_count = r300->velems->count;
unsigned i, v, vbi;
/* Size of the vertex, in dwords. */
unsigned vertex_size = r300->velems->vertex_size_dwords;
/* The number of dwords for this draw operation. */
unsigned dwords = 9 + count * vertex_size;
/* 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];
/* Mapped vertex buffers. */
uint32_t* map[PIPE_MAX_ATTRIBS];
uint32_t* mapelem[PIPE_MAX_ATTRIBS];
struct pipe_transfer* transfer[PIPE_MAX_ATTRIBS] = {0};
CS_LOCALS(r300);
if (!r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0))
return;
/* Calculate the vertex size, offsets, strides etc. and map the buffers. */
for (i = 0; i < vertex_element_count; i++) {
velem = &r300->velems->velem[i];
size[i] = r300->velems->hw_format_size[i] / 4;
vbi = velem->vertex_buffer_index;
vbuf = &r300->vertex_buffer[vbi];
stride[i] = vbuf->stride / 4;
/* Map the buffer. */
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[i] * start;
}
mapelem[i] = map[vbi] + (velem->src_offset / 4);
}
BEGIN_CS(dwords);
OUT_CS_REG(R300_GA_COLOR_CONTROL,
r300_provoking_vertex_fixes(r300, mode));
OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
OUT_CS(count - 1);
OUT_CS(0);
OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, count * vertex_size);
OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (count << 16) |
r300_translate_primitive(mode));
/* Emit vertices. */
for (v = 0; v < count; v++) {
for (i = 0; i < vertex_element_count; i++) {
OUT_CS_TABLE(&mapelem[i][stride[i] * v], size[i]);
}
}
END_CS;
/* Unmap buffers. */
for (i = 0; i < vertex_element_count; i++) {
vbi = r300->velems->velem[i].vertex_buffer_index;
if (transfer[vbi]) {
vbuf = &r300->vertex_buffer[vbi];
pipe_buffer_unmap(&r300->context, vbuf->buffer, transfer[vbi]);
transfer[vbi] = NULL;
}
}
}
/* This functions is used to draw a rectangle for the blitter module.
*
* If we rendered a quad, 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;
if (!r300_prepare_for_rendering(r300, PREP_FIRST_DRAW, NULL, dwords, 0, 0))
goto done;
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;
done:
/* 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;
}
