#include "util/u_format.h"
#include "util/u_math.h"
+#include "util/u_mm.h"
#include "util/u_simple_list.h"
#include "r300_context.h"
{
struct r300_textures_state* texstate = r300->textures_state.state;
static float vec[4] = { 0.0, 0.0, 0.0, 1.0 };
- struct pipe_resource *tex;
+ struct r300_texture *tex;
assert(constant->Type == RC_CONSTANT_STATE);
/* Factor for converting rectangle coords to
* normalized coords. Should only show up on non-r500. */
case RC_STATE_R300_TEXRECT_FACTOR:
- tex = texstate->sampler_views[constant->u.State[1]]->base.texture;
- vec[0] = 1.0 / tex->width0;
- vec[1] = 1.0 / tex->height0;
+ tex = r300_texture(texstate->sampler_views[constant->u.State[1]]->base.texture);
+ vec[0] = 1.0 / tex->desc.width0;
+ vec[1] = 1.0 / tex->desc.height0;
+ break;
+
+ case RC_STATE_R300_TEXSCALE_FACTOR:
+ tex = r300_texture(texstate->sampler_views[constant->u.State[1]]->base.texture);
+ /* Add a small number to the texture size to work around rounding errors in hw. */
+ vec[0] = tex->desc.b.b.width0 / (tex->desc.width0 + 0.001f);
+ vec[1] = tex->desc.b.b.height0 / (tex->desc.height0 + 0.001f);
+ vec[2] = tex->desc.b.b.depth0 / (tex->desc.depth0 + 0.001f);
break;
case RC_STATE_R300_VIEWPORT_SCALE:
BEGIN_CS(size);
OUT_CS_REG_SEQ(R300_PFS_PARAM_0_X, count * 4);
- for (i = 0; i < count; i++)
- for (j = 0; j < 4; j++)
- OUT_CS(pack_float24(buf->ptr[i*4+j]));
+ if (buf->remap_table){
+ for (i = 0; i < count; i++) {
+ float *data = (float*)&buf->ptr[buf->remap_table[i]*4];
+ for (j = 0; j < 4; j++)
+ OUT_CS(pack_float24(data[j]));
+ }
+ } else {
+ for (i = 0; i < count; i++)
+ for (j = 0; j < 4; j++)
+ OUT_CS(pack_float24(*(float*)&buf->ptr[i*4+j]));
+ }
+
END_CS;
}
{
struct r300_fragment_shader *fs = r300_fs(r300);
struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state;
- unsigned count = fs->shader->externals_count * 4;
+ unsigned count = fs->shader->externals_count;
CS_LOCALS(r300);
if (count == 0)
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);
- OUT_CS_TABLE(buf->ptr, count);
+ 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;
}
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.
struct pipe_framebuffer_state* fb = (struct pipe_framebuffer_state*)state;
struct r300_surface* surf;
unsigned i;
+ boolean can_hyperz = r300->rws->get_value(r300->rws, R300_CAN_HYPERZ);
CS_LOCALS(r300);
BEGIN_CS(size);
OUT_CS_REG_SEQ(R300_ZB_DEPTHPITCH, 1);
OUT_CS_RELOC(surf->buffer, surf->cbzb_pitch, 0, surf->domain);
+
+ DBG(r300, DBG_CBZB,
+ "CBZB clearing cbuf %08x %08x\n", surf->cbzb_format,
+ surf->cbzb_pitch);
}
/* Set up a zbuffer. */
else if (fb->zsbuf) {
OUT_CS_REG_SEQ(R300_ZB_DEPTHPITCH, 1);
OUT_CS_RELOC(surf->buffer, surf->pitch, 0, surf->domain);
- /* HiZ RAM. */
- if (r300->screen->caps.has_hiz) {
- OUT_CS_REG(R300_ZB_HIZ_OFFSET, 0);
- OUT_CS_REG(R300_ZB_HIZ_PITCH, 0);
+ if (can_hyperz) {
+ uint32_t surf_pitch;
+ struct r300_texture *tex;
+ int level = surf->base.level;
+ tex = r300_texture(surf->base.texture);
+
+ surf_pitch = surf->pitch & R300_DEPTHPITCH_MASK;
+ /* HiZ RAM. */
+ if (r300->screen->caps.hiz_ram) {
+ if (tex->hiz_mem[level]) {
+ OUT_CS_REG(R300_ZB_HIZ_OFFSET, tex->hiz_mem[level]->ofs << 2);
+ OUT_CS_REG(R300_ZB_HIZ_PITCH, surf_pitch);
+ } else {
+ OUT_CS_REG(R300_ZB_HIZ_OFFSET, 0);
+ OUT_CS_REG(R300_ZB_HIZ_PITCH, 0);
+ }
+ }
+ /* Z Mask RAM. (compressed zbuffer) */
+ if (tex->zmask_mem[level]) {
+ OUT_CS_REG(R300_ZB_ZMASK_OFFSET, tex->zmask_mem[level]->ofs << 2);
+ OUT_CS_REG(R300_ZB_ZMASK_PITCH, surf_pitch);
+ } else {
+ OUT_CS_REG(R300_ZB_ZMASK_OFFSET, 0);
+ OUT_CS_REG(R300_ZB_ZMASK_PITCH, 0);
+ }
}
-
- /* Z Mask RAM. (compressed zbuffer) */
- OUT_CS_REG(R300_ZB_ZMASK_OFFSET, 0);
- OUT_CS_REG(R300_ZB_ZMASK_PITCH, 0);
}
END_CS;
void r300_emit_hyperz_state(struct r300_context *r300,
unsigned size, void *state)
{
+ struct r300_hyperz_state *z = state;
CS_LOCALS(r300);
- WRITE_CS_TABLE(state, size);
+ if (z->flush)
+ WRITE_CS_TABLE(&z->cb_flush_begin, size);
+ else
+ WRITE_CS_TABLE(&z->cb_begin, size - 2);
}
void r300_emit_hyperz_end(struct r300_context *r300)
struct r300_hyperz_state z =
*(struct r300_hyperz_state*)r300->hyperz_state.state;
+ z.flush = 1;
z.zb_bw_cntl = 0;
z.zb_depthclearvalue = 0;
z.sc_hyperz = R300_SC_HYPERZ_ADJ_2;
+ z.gb_z_peq_config = 0;
r300_emit_hyperz_state(r300, r300->hyperz_state.size, &z);
}
CS_LOCALS(r300);
BEGIN_CS(size);
- OUT_CS_TABLE(rs->cb_main, 25);
+ 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);
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);
OUT_CS(1 | (!indexed ? R300_VC_FORCE_PREFETCH : 0));
OUT_CS(r300->vertex_info.size |
(r300->vertex_info.size << 8));
- OUT_CS(r300->vbo_offset);
+ OUT_CS(r300->draw_vbo_offset);
OUT_CS_BUF_RELOC(r300->vbo, 0, r300_buffer(r300->vbo)->domain, 0);
END_CS;
}
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, 6);
+ unsigned pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 5);
unsigned imm_first = vs->externals_count;
unsigned imm_end = vs->code.constants.Count;
OUT_CS_TABLE(data, 4);
}
}
+
+ /* Emit flow control instructions. */
+ if (code->num_fc_ops) {
+
+ 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, code->num_fc_ops * 2);
+ OUT_CS_TABLE(code->fc_op_addrs.r500, code->num_fc_ops * 2);
+ } else {
+ OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_ADDRS_0, code->num_fc_ops);
+ OUT_CS_TABLE(code->fc_op_addrs.r300, code->num_fc_ops);
+ }
+ OUT_CS_REG_SEQ(R300_VAP_PVS_FLOW_CNTL_LOOP_INDEX_0, code->num_fc_ops);
+ OUT_CS_TABLE(code->fc_loop_index, code->num_fc_ops);
+ }
+
END_CS;
}
unsigned count =
((struct r300_vertex_shader*)r300->vs_state.state)->externals_count;
struct r300_constant_buffer *buf = (struct r300_constant_buffer*)state;
+ unsigned i;
CS_LOCALS(r300);
if (!count)
(r300->screen->caps.is_r500 ?
R500_PVS_CONST_START : R300_PVS_CONST_START));
OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, count * 4);
- OUT_CS_TABLE(buf->ptr, 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);
+ }
END_CS;
}
END_CS;
}
+static void r300_emit_hiz_line_clear(struct r300_context *r300, int start, uint16_t count, uint32_t val)
+{
+ CS_LOCALS(r300);
+ BEGIN_CS(4);
+ OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_HIZ, 2);
+ OUT_CS(start);
+ OUT_CS(count);
+ OUT_CS(val);
+ END_CS;
+}
+
+static void r300_emit_zmask_line_clear(struct r300_context *r300, int start, uint16_t count, uint32_t val)
+{
+ CS_LOCALS(r300);
+ BEGIN_CS(4);
+ OUT_CS_PKT3(R300_PACKET3_3D_CLEAR_ZMASK, 2);
+ OUT_CS(start);
+ OUT_CS(count);
+ OUT_CS(val);
+ END_CS;
+}
+
+#define ALIGN_DIVUP(x, y) (((x) + (y) - 1) / (y))
+
+void r300_emit_hiz_clear(struct r300_context *r300, unsigned size, void *state)
+{
+ struct pipe_framebuffer_state *fb =
+ (struct pipe_framebuffer_state*)r300->fb_state.state;
+ struct r300_hyperz_state *z =
+ (struct r300_hyperz_state*)r300->hyperz_state.state;
+ struct r300_screen* r300screen = r300->screen;
+ uint32_t stride, offset = 0, height, offset_shift;
+ struct r300_texture* tex;
+ int i;
+
+ tex = r300_texture(fb->zsbuf->texture);
+
+ offset = tex->hiz_mem[fb->zsbuf->level]->ofs;
+ stride = tex->desc.stride_in_pixels[fb->zsbuf->level];
+
+ /* convert from pixels to 4x4 blocks */
+ stride = ALIGN_DIVUP(stride, 4);
+
+ stride = ALIGN_DIVUP(stride, r300screen->caps.num_frag_pipes);
+ /* there are 4 blocks per dwords */
+ stride = ALIGN_DIVUP(stride, 4);
+
+ height = ALIGN_DIVUP(fb->zsbuf->height, 4);
+
+ offset_shift = 2;
+ offset_shift += (r300screen->caps.num_frag_pipes / 2);
+
+ for (i = 0; i < height; i++) {
+ offset = i * stride;
+ offset <<= offset_shift;
+ r300_emit_hiz_line_clear(r300, offset, stride, 0xffffffff);
+ }
+ z->current_func = -1;
+
+ /* Mark the current zbuffer's hiz ram as in use. */
+ tex->hiz_in_use[fb->zsbuf->level] = TRUE;
+}
+
+void r300_emit_zmask_clear(struct r300_context *r300, unsigned size, void *state)
+{
+ struct pipe_framebuffer_state *fb =
+ (struct pipe_framebuffer_state*)r300->fb_state.state;
+ struct r300_screen* r300screen = r300->screen;
+ uint32_t stride, offset = 0;
+ struct r300_texture* tex;
+ uint32_t i, height;
+ int mult, offset_shift;
+
+ tex = r300_texture(fb->zsbuf->texture);
+ stride = tex->desc.stride_in_pixels[fb->zsbuf->level];
+
+ offset = tex->zmask_mem[fb->zsbuf->level]->ofs;
+
+ if (r300->z_compression == RV350_Z_COMPRESS_88)
+ mult = 8;
+ else
+ mult = 4;
+
+ height = ALIGN_DIVUP(fb->zsbuf->height, mult);
+
+ offset_shift = 4;
+ offset_shift += (r300screen->caps.num_frag_pipes / 2);
+ stride = ALIGN_DIVUP(stride, r300screen->caps.num_frag_pipes);
+
+ /* okay have width in pixels - divide by block width */
+ stride = ALIGN_DIVUP(stride, mult);
+ /* have width in blocks - divide by number of fragment pipes screen width */
+ /* 16 blocks per dword */
+ stride = ALIGN_DIVUP(stride, 16);
+
+ for (i = 0; i < height; i++) {
+ offset = i * stride;
+ offset <<= offset_shift;
+ r300_emit_zmask_line_clear(r300, offset, stride, 0x0);//0xffffffff);
+ }
+
+ /* Mark the current zbuffer's zmask as in use. */
+ tex->zmask_in_use[fb->zsbuf->level] = TRUE;
+}
+
void r300_emit_ztop_state(struct r300_context* r300,
unsigned size, void* state)
{
END_CS;
}
-void r300_emit_buffer_validate(struct r300_context *r300,
- boolean do_validate_vertex_buffers,
- struct pipe_resource *index_buffer)
+boolean r300_emit_buffer_validate(struct r300_context *r300,
+ boolean do_validate_vertex_buffers,
+ struct pipe_resource *index_buffer)
{
struct pipe_framebuffer_state* fb =
(struct pipe_framebuffer_state*)r300->fb_state.state;
struct pipe_vertex_element *velem = r300->velems->velem;
struct pipe_resource *pbuf;
unsigned i;
- boolean invalid = FALSE;
/* upload buffers first */
if (r300->screen->caps.has_tcl && r300->any_user_vbs) {
/* Clean out BOs. */
r300->rws->cs_reset_buffers(r300->cs);
-validate:
/* Color buffers... */
for (i = 0; i < fb->nr_cbufs; i++) {
tex = r300_texture(fb->cbufs[i]->texture);
r300_buffer(index_buffer)->domain, 0);
if (!r300->rws->cs_validate(r300->cs)) {
- r300->context.flush(&r300->context, 0, NULL);
- if (invalid) {
- /* Well, hell. */
- fprintf(stderr, "r300: Stuck in validation loop, gonna quit now.\n");
- abort();
- }
- invalid = TRUE;
- goto validate;
+ return FALSE;
}
+
+ return TRUE;
}
unsigned r300_get_num_dirty_dwords(struct r300_context *r300)
return dwords;
}
+unsigned r300_get_num_cs_end_dwords(struct r300_context *r300)
+{
+ unsigned dwords = 0;
+
+ /* Emitted in flush. */
+ dwords += 26; /* emit_query_end */
+ dwords += r300->hyperz_state.size + 2; /* emit_hyperz_end + zcache flush */
+ if (r500_index_bias_supported(r300))
+ dwords += 2;
+
+ return dwords;
+}
+
/* Emit all dirty state. */
void r300_emit_dirty_state(struct r300_context* r300)
{