FREE(r300->rs_block_state.state);
FREE(r300->scissor_state.state);
FREE(r300->textures_state.state);
- FREE(r300->vap_output_state.state);
FREE(r300->viewport_state.state);
FREE(r300->ztop_state.state);
FREE(r300->fs_constants.state);
R300_INIT_ATOM(viewport_state, 9);
R300_INIT_ATOM(rs_block_state, 0);
R300_INIT_ATOM(vertex_stream_state, 0);
- R300_INIT_ATOM(vap_output_state, 6);
R300_INIT_ATOM(pvs_flush, 2);
R300_INIT_ATOM(vs_state, 0);
R300_INIT_ATOM(vs_constants, 0);
r300->rs_block_state.state = CALLOC_STRUCT(r300_rs_block);
r300->scissor_state.state = CALLOC_STRUCT(pipe_scissor_state);
r300->textures_state.state = CALLOC_STRUCT(r300_textures_state);
- r300->vap_output_state.state = CALLOC_STRUCT(r300_vap_output_state);
r300->viewport_state.state = CALLOC_STRUCT(r300_viewport_state);
r300->ztop_state.state = CALLOC_STRUCT(r300_ztop_state);
r300->fs_constants.state = CALLOC_STRUCT(r300_constant_buffer);
};
struct r300_rs_block {
+ uint32_t vap_vtx_state_cntl; /* R300_VAP_VTX_STATE_CNTL: 0x2180 */
+ uint32_t vap_vsm_vtx_assm; /* R300_VAP_VSM_VTX_ASSM: 0x2184 */
+ uint32_t vap_out_vtx_fmt[2]; /* R300_VAP_OUTPUT_VTX_FMT_[0-1]: 0x2090 */
+
uint32_t ip[8]; /* R300_RS_IP_[0-7], R500_RS_IP_[0-7] */
uint32_t count; /* R300_RS_COUNT */
uint32_t inst_count; /* R300_RS_INST_COUNT */
unsigned count;
};
-struct r300_vap_output_state {
- uint32_t vap_vtx_state_cntl; /* R300_VAP_VTX_STATE_CNTL: 0x2180 */
- uint32_t vap_vsm_vtx_assm; /* R300_VAP_VSM_VTX_ASSM: 0x2184 */
- uint32_t vap_out_vtx_fmt[2]; /* R300_VAP_OUTPUT_VTX_FMT_[0-1]: 0x2090 */
-};
-
struct r300_viewport_state {
float xscale; /* R300_VAP_VPORT_XSCALE: 0x2098 */
float xoffset; /* R300_VAP_VPORT_XOFFSET: 0x209c */
struct r300_atom query_start;
/* Rasterizer state. */
struct r300_atom rs_state;
- /* RS block state. */
+ /* RS block state + VAP (vertex shader) output mapping state. */
struct r300_atom rs_block_state;
/* Scissor state. */
struct r300_atom scissor_state;
struct r300_atom textures_state;
/* Vertex stream formatting state. */
struct r300_atom vertex_stream_state;
- /* VAP (vertex shader) output mapping state. */
- struct r300_atom vap_output_state;
/* Vertex shader. */
struct r300_atom vs_state;
/* Vertex shader constant buffer. */
struct pipe_viewport_state viewport;
+ /* Stream locations for SWTCL. */
+ int stream_loc_notcl[16];
+
/* Flag indicating whether or not the HW is dirty. */
uint32_t dirty_hw;
/* Whether polygon offset is enabled. */
DBG(r300, DBG_DRAW, "r300: RS emit:\n");
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]);
+
if (r300->screen->caps.is_r500) {
OUT_CS_REG_SEQ(R500_RS_IP_0, count);
} else {
END_CS;
}
-void r300_emit_vap_output_state(struct r300_context* r300,
- unsigned size, void* state)
-{
- struct r300_vap_output_state *vap_out_state =
- (struct r300_vap_output_state*)state;
- CS_LOCALS(r300);
-
- DBG(r300, DBG_DRAW, "r300: VAP emit:\n");
-
- BEGIN_CS(size);
- OUT_CS_REG_SEQ(R300_VAP_VTX_STATE_CNTL, 2);
- OUT_CS(vap_out_state->vap_vtx_state_cntl);
- OUT_CS(vap_out_state->vap_vsm_vtx_assm);
- OUT_CS_REG_SEQ(R300_VAP_OUTPUT_VTX_FMT_0, 2);
- OUT_CS(vap_out_state->vap_out_vtx_fmt[0]);
- OUT_CS(vap_out_state->vap_out_vtx_fmt[1]);
- END_CS;
-}
-
void r300_emit_pvs_flush(struct r300_context* r300, unsigned size, void* state)
{
CS_LOCALS(r300);
void r300_emit_vertex_stream_state(struct r300_context* r300,
unsigned size, void* state);
-void r300_emit_vap_output_state(struct r300_context* r300,
- unsigned size, void* state);
-
void r300_emit_vs_constants(struct r300_context* r300,
unsigned size, void *state);
r300_mark_fs_code_dirty(r300);
r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */
-
- if (r300->vs_state.state && r300_vertex_shader_setup_wpos(r300)) {
- r300->vap_output_state.dirty = TRUE;
- }
}
/* Delete fragment shader state. */
}
r300->vs_state.state = vs;
- // VS output mapping for HWTCL or stream mapping for SWTCL to the RS block
- if (r300->fs.state) {
- r300_vertex_shader_setup_wpos(r300);
- }
- memcpy(r300->vap_output_state.state, &vs->vap_out,
- sizeof(struct r300_vap_output_state));
- r300->vap_output_state.dirty = TRUE;
-
/* The majority of the RS block bits is dependent on the vertex shader. */
r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */
static void r300_swtcl_vertex_psc(struct r300_context *r300)
{
struct r300_vertex_stream_state *vstream = r300->vertex_stream_state.state;
- struct r300_vertex_shader* vs = r300->vs_state.state;
- struct vertex_info* vinfo = &r300->vertex_info;
+ struct vertex_info* vinfo = &r300->vertex_info;
uint16_t type, swizzle;
enum pipe_format format;
unsigned i, attrib_count;
- int* vs_output_tab = vs->stream_loc_notcl;
+ int* vs_output_tab = r300->stream_loc_notcl;
/* XXX hax */
memset(vstream, 0, sizeof(struct r300_vertex_stream_state));
/* Set up the RS block.
*
- * This is the part of the chipset that actually does the rasterization
- * of vertices into fragments. This is also the part of the chipset that
- * locks up if any part of it is even slightly wrong. */
+ * This is the part of the chipset that is responsible for linking vertex
+ * and fragment shaders and stuffed texture coordinates.
+ *
+ * The rasterizer reads data from VAP, which produces vertex shader outputs,
+ * and GA, which produces stuffed texture coordinates. VAP outputs have
+ * precedence over GA. All outputs must be rasterized otherwise it locks up.
+ * If there are more outputs rasterized than is set in VAP/GA, it locks up
+ * too. The funky part is that this info has been pretty much obtained by trial
+ * and error. */
static void r300_update_rs_block(struct r300_context *r300)
{
struct r300_vertex_shader *vs = r300->vs_state.state;
struct r300_shader_semantics *vs_outputs = &vs->outputs;
struct r300_shader_semantics *fs_inputs = &r300_fs(r300)->shader->inputs;
- struct r300_rs_block rs = { { 0 } };
- int i, col_count = 0, tex_count = 0, fp_offset = 0, count;
+ struct r300_rs_block rs = {0};
+ int i, col_count = 0, tex_count = 0, fp_offset = 0, count, loc = 0;
void (*rX00_rs_col)(struct r300_rs_block*, int, int, enum r300_rs_swizzle);
void (*rX00_rs_col_write)(struct r300_rs_block*, int, int);
void (*rX00_rs_tex)(struct r300_rs_block*, int, int, enum r300_rs_swizzle);
void (*rX00_rs_tex_write)(struct r300_rs_block*, int, int);
boolean any_bcolor_used = vs_outputs->bcolor[0] != ATTR_UNUSED ||
vs_outputs->bcolor[1] != ATTR_UNUSED;
+ int *stream_loc_notcl = r300->stream_loc_notcl;
if (r300->screen->caps.is_r500) {
rX00_rs_col = r500_rs_col;
rX00_rs_tex_write = r300_rs_tex_write;
}
- /* Rasterize colors. */
+ /* The position is always present in VAP. */
+ rs.vap_vsm_vtx_assm |= R300_INPUT_CNTL_POS;
+ rs.vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
+ stream_loc_notcl[loc++] = 0;
+
+ /* Set up the point size in VAP. */
+ if (vs_outputs->psize != ATTR_UNUSED) {
+ rs.vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT;
+ stream_loc_notcl[loc++] = 1;
+ }
+
+ /* Set up and rasterize colors. */
for (i = 0; i < ATTR_COLOR_COUNT; i++) {
if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used ||
vs_outputs->color[1] != ATTR_UNUSED) {
- /* Always rasterize if it's written by the VS,
- * otherwise it locks up. */
+ /* Set up the color in VAP. */
+ rs.vap_vsm_vtx_assm |= R300_INPUT_CNTL_COLOR;
+ rs.vap_out_vtx_fmt[0] |=
+ R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i;
+ stream_loc_notcl[loc++] = 2 + i;
+
+ /* Rasterize it. */
rX00_rs_col(&rs, col_count, col_count, SWIZ_XYZW);
- /* Write it to the FS input register if it's used by the FS. */
+ /* Write it to the FS input register if it's needed by the FS. */
if (fs_inputs->color[i] != ATTR_UNUSED) {
rX00_rs_col_write(&rs, col_count, fp_offset);
fp_offset++;
}
}
+ /* Set up back-face colors. The rasterizer will do the color selection
+ * automatically. */
+ if (any_bcolor_used) {
+ for (i = 0; i < ATTR_COLOR_COUNT; i++) {
+ rs.vap_vsm_vtx_assm |= R300_INPUT_CNTL_COLOR;
+ rs.vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << (2+i);
+ stream_loc_notcl[loc++] = 4 + i;
+ }
+ }
+
/* Rasterize texture coordinates. */
- for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
+ for (i = 0; i < ATTR_GENERIC_COUNT && tex_count < 8; i++) {
bool sprite_coord = !!(r300->sprite_coord_enable & (1 << i));
if (vs_outputs->generic[i] != ATTR_UNUSED || sprite_coord) {
- /* Always rasterize if it's written by the VS,
- * otherwise it locks up. */
+ if (!sprite_coord) {
+ /* Set up the texture coordinates in VAP. */
+ rs.vap_vsm_vtx_assm |= (R300_INPUT_CNTL_TC0 << tex_count);
+ rs.vap_out_vtx_fmt[1] |= (4 << (3 * tex_count));
+ stream_loc_notcl[loc++] = 6 + tex_count;
+ }
+
+ /* Rasterize it. */
rX00_rs_tex(&rs, tex_count, tex_count,
sprite_coord ? SWIZ_XY01 : SWIZ_XYZW);
- /* Write it to the FS input register if it's used by the FS. */
+ /* Write it to the FS input register if it's needed by the FS. */
if (fs_inputs->generic[i] != ATTR_UNUSED) {
rX00_rs_tex_write(&rs, tex_count, fp_offset);
fp_offset++;
}
/* Rasterize fog coordinates. */
- if (vs_outputs->fog != ATTR_UNUSED) {
- /* Always rasterize if it's written by the VS,
- * otherwise it locks up. */
+ if (vs_outputs->fog != ATTR_UNUSED && tex_count < 8) {
+ /* Set up the fog coordinates in VAP. */
+ rs.vap_vsm_vtx_assm |= (R300_INPUT_CNTL_TC0 << tex_count);
+ rs.vap_out_vtx_fmt[1] |= (4 << (3 * tex_count));
+ stream_loc_notcl[loc++] = 6 + tex_count;
+
+ /* Rasterize it. */
rX00_rs_tex(&rs, tex_count, tex_count, SWIZ_X001);
- /* Write it to the FS input register if it's used by the FS. */
+ /* Write it to the FS input register if it's needed by the FS. */
if (fs_inputs->fog != ATTR_UNUSED) {
rX00_rs_tex_write(&rs, tex_count, fp_offset);
fp_offset++;
}
/* Rasterize WPOS. */
- /* If the FS doesn't need it, it's not written by the VS. */
- if (vs_outputs->wpos != ATTR_UNUSED && fs_inputs->wpos != ATTR_UNUSED) {
+ /* Don't set it in VAP if the FS doesn't need it. */
+ if (fs_inputs->wpos != ATTR_UNUSED && tex_count < 8) {
+ /* Set up the WPOS coordinates in VAP. */
+ rs.vap_vsm_vtx_assm |= (R300_INPUT_CNTL_TC0 << tex_count);
+ rs.vap_out_vtx_fmt[1] |= (4 << (3 * tex_count));
+ stream_loc_notcl[loc++] = 6 + tex_count;
+
+ /* Rasterize it. */
rX00_rs_tex(&rs, tex_count, tex_count, SWIZ_XYZW);
+
+ /* Write it to the FS input register. */
rX00_rs_tex_write(&rs, tex_count, fp_offset);
DBG(r300, DBG_RS, "r300: Rasterized WPOS written to FS.\n");
tex_count++;
}
+ /* Invalidate the rest of the no-TCL (GA) stream locations. */
+ for (; loc < 16;) {
+ stream_loc_notcl[loc++] = -1;
+ }
+
/* Rasterize at least one color, or bad things happen. */
if (col_count == 0 && tex_count == 0) {
rX00_rs_col(&rs, 0, 0, SWIZ_0001);
/* Now, after all that, see if we actually need to update the state. */
if (memcmp(r300->rs_block_state.state, &rs, sizeof(struct r300_rs_block))) {
memcpy(r300->rs_block_state.state, &rs, sizeof(struct r300_rs_block));
- r300->rs_block_state.size = 5 + count*2;
+ r300->rs_block_state.size = 11 + count*2;
}
}
vs_outputs->wpos = i;
}
-/* This function sets up:
- * - VAP mapping, which maps VS registers to output semantics and
- * at the same time it indicates which attributes are enabled and should
- * be rasterized.
- * - Stream mapping to VS outputs if TCL is not present. */
-static void r300_init_vs_output_mapping(struct r300_vertex_shader* vs)
-{
- struct r300_shader_semantics* vs_outputs = &vs->outputs;
- struct r300_vap_output_state *vap_out = &vs->vap_out;
- int *stream_loc = vs->stream_loc_notcl;
- int i, gen_count, tabi = 0;
- boolean any_bcolor_used = vs_outputs->bcolor[0] != ATTR_UNUSED ||
- vs_outputs->bcolor[1] != ATTR_UNUSED;
-
- vap_out->vap_vtx_state_cntl = 0x5555; /* XXX this is classic Mesa bonghits */
-
- /* Position. */
- if (vs_outputs->pos != ATTR_UNUSED) {
- vap_out->vap_vsm_vtx_assm |= R300_INPUT_CNTL_POS;
- vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
-
- stream_loc[tabi++] = 0;
- } else {
- assert(0);
- }
-
- /* Point size. */
- if (vs_outputs->psize != ATTR_UNUSED) {
- vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT;
-
- stream_loc[tabi++] = 1;
- }
-
- /* Colors. */
- for (i = 0; i < ATTR_COLOR_COUNT; i++) {
- if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used ||
- vs_outputs->color[1] != ATTR_UNUSED) {
- vap_out->vap_vsm_vtx_assm |= R300_INPUT_CNTL_COLOR;
- vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i;
-
- stream_loc[tabi++] = 2 + i;
- }
- }
-
- /* Back-face colors. */
- if (any_bcolor_used) {
- for (i = 0; i < ATTR_COLOR_COUNT; i++) {
- vap_out->vap_vsm_vtx_assm |= R300_INPUT_CNTL_COLOR;
- vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << (2+i);
-
- stream_loc[tabi++] = 4 + i;
- }
- }
-
- /* Texture coordinates. */
- gen_count = 0;
- for (i = 0; i < ATTR_GENERIC_COUNT && gen_count < 8; i++) {
- if (vs_outputs->generic[i] != ATTR_UNUSED) {
- vap_out->vap_vsm_vtx_assm |= (R300_INPUT_CNTL_TC0 << gen_count);
- vap_out->vap_out_vtx_fmt[1] |= (4 << (3 * gen_count));
-
- stream_loc[tabi++] = 6 + gen_count;
- gen_count++;
- }
- }
-
- /* Fog coordinates. */
- if (gen_count < 8 && vs_outputs->fog != ATTR_UNUSED) {
- vap_out->vap_vsm_vtx_assm |= (R300_INPUT_CNTL_TC0 << gen_count);
- vap_out->vap_out_vtx_fmt[1] |= (4 << (3 * gen_count));
-
- stream_loc[tabi++] = 6 + gen_count;
- gen_count++;
- }
-
- /* WPOS. */
- if (gen_count < 8) {
- vs->wpos_tex_output = gen_count;
- stream_loc[tabi++] = 6 + gen_count;
- } else {
- vs_outputs->wpos = ATTR_UNUSED;
- }
-
- for (; tabi < 16;) {
- stream_loc[tabi++] = -1;
- }
-}
-
static void set_vertex_inputs_outputs(struct r300_vertex_program_compiler * c)
{
struct r300_vertex_shader * vs = c->UserData;
}
/* WPOS. */
- if (outputs->wpos != ATTR_UNUSED) {
- c->code->outputs[outputs->wpos] = reg++;
- }
+ c->code->outputs[outputs->wpos] = reg++;
}
static void r300_dummy_vertex_shader(
tgsi_scan_shader(tokens, &vs->info);
r300_shader_read_vs_outputs(&vs->info, &vs->outputs);
- r300_init_vs_output_mapping(vs);
/* Setup the compiler */
rc_init(&compiler.Base);
r300_tgsi_to_rc(&ttr, tokens);
- compiler.RequiredOutputs =
- ~(~0 << (vs->info.num_outputs +
- (vs->outputs.wpos != ATTR_UNUSED ? 1 : 0)));
-
+ compiler.RequiredOutputs = ~(~0 << (vs->info.num_outputs + 1));
compiler.SetHwInputOutput = &set_vertex_inputs_outputs;
/* Insert the WPOS output. */
- if (vs->outputs.wpos != ATTR_UNUSED) {
- rc_copy_output(&compiler.Base, 0, vs->outputs.wpos);
- }
+ rc_copy_output(&compiler.Base, 0, vs->outputs.wpos);
/* Invoke the compiler */
r3xx_compile_vertex_program(&compiler);
/* And, finally... */
rc_destroy(&compiler.Base);
}
-
-boolean r300_vertex_shader_setup_wpos(struct r300_context* r300)
-{
- struct r300_vertex_shader* vs = r300->vs_state.state;
- struct r300_vap_output_state *vap_out = &vs->vap_out;
- int tex_output = vs->wpos_tex_output;
- uint32_t tex_fmt = R300_INPUT_CNTL_TC0 << tex_output;
-
- if (vs->outputs.wpos == ATTR_UNUSED) {
- return FALSE;
- }
-
- if (r300_fs(r300)->shader->inputs.wpos != ATTR_UNUSED) {
- /* Enable WPOS in VAP. */
- if (!(vap_out->vap_vsm_vtx_assm & tex_fmt)) {
- vap_out->vap_vsm_vtx_assm |= tex_fmt;
- vap_out->vap_out_vtx_fmt[1] |= (4 << (3 * tex_output));
- return TRUE;
- }
- } else {
- /* Disable WPOS in VAP. */
- if (vap_out->vap_vsm_vtx_assm & tex_fmt) {
- vap_out->vap_vsm_vtx_assm &= ~tex_fmt;
- vap_out->vap_out_vtx_fmt[1] &= ~(4 << (3 * tex_output));
- return TRUE;
- }
- }
- return FALSE;
-}
struct tgsi_shader_info info;
struct r300_shader_semantics outputs;
- struct r300_vap_output_state vap_out;
/* Whether the shader was replaced by a dummy one due to a shader
* compilation failure. */
unsigned externals_count;
unsigned immediates_count;
- /* Stream locations for SWTCL or if TCL is bypassed. */
- int stream_loc_notcl[16];
-
- /* Output stream location for WPOS. */
- int wpos_tex_output;
-
/* HWTCL-specific. */
/* Machine code (if translated) */
struct r300_vertex_program_code code;
struct r300_vertex_shader* vs,
const struct tgsi_token *tokens);
-/* Return TRUE if VAP (hwfmt) needs to be re-emitted. */
-boolean r300_vertex_shader_setup_wpos(struct r300_context* r300);
-
#endif /* R300_VS_H */
projects / mesa.git / commitdiff