r300_state_derived.c revision 2305642b2e8edcebdc727f1181f7dbfcc78e8028 
1/*
2 * Copyright 2008 Corbin Simpson <MostAwesomeDude@gmail.com>
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * on the rights to use, copy, modify, merge, publish, distribute, sub
8 * license, and/or sell copies of the Software, and to permit persons to whom
9 * the Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice (including the next
12 * paragraph) shall be included in all copies or substantial portions of the
13 * Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
18 * THE AUTHOR(S) AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
19 * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
20 * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
21 * USE OR OTHER DEALINGS IN THE SOFTWARE. */
22
23#include "r300_state_derived.h"
24
25/* r300_state_derived: Various bits of state which are dependent upon
26 * currently bound CSO data. */
27
28/* Update the vertex_info struct in our r300_context.
29 *
30 * The vertex_info struct describes the post-TCL format of vertices. It is
31 * required for Draw when doing SW TCL, and also for describing the
32 * dreaded RS block on R300 chipsets. */
33/* XXX this function should be able to handle vert shaders as well as draw */
34static void r300_update_vertex_layout(struct r300_context* r300)
35{
36    struct r300_vertex_format vformat;
37    struct vertex_info vinfo;
38    boolean pos = FALSE, psize = FALSE, fog = FALSE;
39    int i, texs = 0, cols = 0;
40    int tab[16];
41
42    struct tgsi_shader_info* info = &r300->fs->info;
43
44    memset(&vinfo, 0, sizeof(vinfo));
45    for (i = 0; i < 16; i++) {
46        tab[i] = -1;
47    }
48
49    assert(info->num_inputs <= 16);
50
51    for (i = 0; i < info->num_inputs; i++) {
52        switch (info->input_semantic_name[i]) {
53            case TGSI_SEMANTIC_POSITION:
54                pos = TRUE;
55                tab[i] = 0;
56                break;
57            case TGSI_SEMANTIC_COLOR:
58                tab[i] = 2 + cols++;
59                break;
60            case TGSI_SEMANTIC_FOG:
61                fog = TRUE;
62                tab[i] = 6 + texs++;
63                break;
64            case TGSI_SEMANTIC_PSIZE:
65                psize = TRUE;
66                tab[i] = 1;
67                break;
68            case TGSI_SEMANTIC_GENERIC:
69                tab[i] = 6 + texs++;
70                break;
71            default:
72                debug_printf("r300: Unknown vertex input %d\n",
73                    info->input_semantic_name[i]);
74                break;
75        }
76    }
77
78    /* Do the actual vertex_info setup.
79     *
80     * vertex_info has four uints of hardware-specific data in it.
81     * vinfo.hwfmt[0] is R300_VAP_VTX_STATE_CNTL
82     * vinfo.hwfmt[1] is R300_VAP_VSM_VTX_ASSM
83     * vinfo.hwfmt[2] is R300_VAP_OUTPUT_VTX_FMT_0
84     * vinfo.hwfmt[3] is R300_VAP_OUTPUT_VTX_FMT_1 */
85
86    vinfo.hwfmt[0] = 0x5555; /* XXX this is classic Mesa bonghits */
87
88    if (!pos) {
89        debug_printf("r300: Forcing vertex position attribute emit...\n");
90        /* Make room for the position attribute
91         * at the beginning of the tab. */
92        for (i = 1; i < 16; i++) {
93            tab[i] = tab[i-1];
94        }
95        tab[0] = 0;
96
97        draw_emit_vertex_attr(&vinfo, EMIT_4F, INTERP_POS,
98            draw_find_vs_output(r300->draw, TGSI_SEMANTIC_POSITION, 0));
99    } else {
100        draw_emit_vertex_attr(&vinfo, EMIT_4F, INTERP_PERSPECTIVE,
101            draw_find_vs_output(r300->draw, TGSI_SEMANTIC_POSITION, 0));
102    }
103    vinfo.hwfmt[1] |= R300_INPUT_CNTL_POS;
104    vinfo.hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
105
106    if (psize) {
107        draw_emit_vertex_attr(&vinfo, EMIT_1F_PSIZE, INTERP_POS,
108            draw_find_vs_output(r300->draw, TGSI_SEMANTIC_PSIZE, 0));
109        vinfo.hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT;
110    }
111
112    for (i = 0; i < cols; i++) {
113        draw_emit_vertex_attr(&vinfo, EMIT_4F, INTERP_LINEAR,
114            draw_find_vs_output(r300->draw, TGSI_SEMANTIC_COLOR, i));
115        vinfo.hwfmt[1] |= R300_INPUT_CNTL_COLOR;
116        vinfo.hwfmt[2] |= (R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i);
117    }
118
119    for (i = 0; i < texs; i++) {
120        draw_emit_vertex_attr(&vinfo, EMIT_4F, INTERP_PERSPECTIVE,
121            draw_find_vs_output(r300->draw, TGSI_SEMANTIC_GENERIC, i));
122        vinfo.hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i);
123        vinfo.hwfmt[3] |= (4 << (3 * i));
124    }
125
126    if (fog) {
127        i++;
128        draw_emit_vertex_attr(&vinfo, EMIT_4F, INTERP_PERSPECTIVE,
129            draw_find_vs_output(r300->draw, TGSI_SEMANTIC_FOG, 0));
130        vinfo.hwfmt[1] |= (R300_INPUT_CNTL_TC0 << i);
131        vinfo.hwfmt[3] |= (4 << (3 * i));
132    }
133
134    draw_compute_vertex_size(&vinfo);
135
136    if (memcmp(&r300->vertex_info, &vinfo, sizeof(struct vertex_info))) {
137        uint32_t temp;
138        debug_printf("attrib count: %d, fp input count: %d\n",
139                vinfo.num_attribs, info->num_inputs);
140        for (i = 0; i < vinfo.num_attribs; i++) {
141            debug_printf("attrib: offset %d, interp %d, size %d,"
142                   " tab %d\n", vinfo.attrib[i].src_index,
143                   vinfo.attrib[i].interp_mode, vinfo.attrib[i].emit,
144                   tab[i]);
145        }
146
147        for (i = 0; i < vinfo.num_attribs; i++) {
148            /* Make sure we have a proper destination for our attribute */
149            assert(tab[i] != -1);
150
151            temp = translate_vertex_data_type(vinfo.attrib[i].emit) |
152                (tab[i] << R300_DST_VEC_LOC_SHIFT);
153            if (i & 1) {
154                r300->vertex_info.vap_prog_stream_cntl[i >> 1] &= 0x0000ffff;
155                r300->vertex_info.vap_prog_stream_cntl[i >> 1] |= temp << 16;
156            } else {
157                r300->vertex_info.vap_prog_stream_cntl[i >> 1] &= 0xffff0000;
158                r300->vertex_info.vap_prog_stream_cntl[i >> 1] |= temp;
159            }
160
161            r300->vertex_info.vap_prog_stream_cntl_ext[i >> 1] |=
162                (R300_VAP_SWIZZLE_XYZW << (i & 1 ? 16 : 0));
163        }
164        /* Set the last vector. */
165        i--;
166        r300->vertex_info.vap_prog_stream_cntl[i >> 1] |= (R300_LAST_VEC <<
167                (i & 1 ? 16 : 0));
168
169        memcpy(r300->vertex_info.tab, tab, sizeof(tab));
170        memcpy(&r300->vertex_info, &vinfo, sizeof(struct vertex_info));
171        r300->dirty_state |= R300_NEW_VERTEX_FORMAT;
172    }
173}
174
175/* Set up the RS block. This is the part of the chipset that actually does
176 * the rasterization of vertices into fragments. This is also the part of the
177 * chipset that locks up if any part of it is even slightly wrong. */
178static void r300_update_rs_block(struct r300_context* r300)
179{
180    struct r300_rs_block* rs = r300->rs_block;
181    struct vertex_info* vinfo = &r300->vertex_info.vinfo;
182    int* tab = r300->vertex_info.tab;
183    int col_count = 0, fp_offset = 0, i, memory_pos, tex_count = 0;
184
185    memset(rs, 0, sizeof(struct r300_rs_block));
186
187    if (r300_screen(r300->context.screen)->caps->is_r500) {
188        for (i = 0; i < vinfo->num_attribs; i++) {
189            assert(tab[vinfo->attrib[i].src_index] != -1);
190            memory_pos = tab[vinfo->attrib[i].src_index] * 4;
191            switch (vinfo->attrib[i].interp_mode) {
192                case INTERP_LINEAR:
193                    rs->ip[col_count] |=
194                        R500_RS_COL_PTR(memory_pos) |
195                        R500_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
196                    col_count++;
197                    break;
198                case INTERP_PERSPECTIVE:
199                    rs->ip[tex_count] |=
200                        R500_RS_SEL_S(memory_pos) |
201                        R500_RS_SEL_T(memory_pos + 1) |
202                        R500_RS_SEL_R(memory_pos + 2) |
203                        R500_RS_SEL_Q(memory_pos + 3);
204                    tex_count++;
205                    break;
206                default:
207                    break;
208            }
209        }
210
211        /* Set up at least one texture pointer or RS will not be happy. */
212        if (tex_count == 0) {
213            rs->ip[0] |=
214                R500_RS_SEL_S(R500_RS_IP_PTR_K0) |
215                R500_RS_SEL_T(R500_RS_IP_PTR_K0) |
216                R500_RS_SEL_R(R500_RS_IP_PTR_K0) |
217                R500_RS_SEL_Q(R500_RS_IP_PTR_K1);
218        }
219
220        for (i = 0; i < tex_count; i++) {
221            rs->inst[i] |= R500_RS_INST_TEX_ID(i) | R500_RS_INST_TEX_CN_WRITE |
222                R500_RS_INST_TEX_ADDR(fp_offset);
223            fp_offset++;
224        }
225
226        for (i = 0; i < col_count; i++) {
227            rs->inst[i] |= R500_RS_INST_COL_ID(i) | R500_RS_INST_COL_CN_WRITE |
228                R500_RS_INST_COL_ADDR(fp_offset);
229            fp_offset++;
230        }
231    } else {
232        for (i = 0; i < vinfo->num_attribs; i++) {
233            memory_pos = tab[vinfo->attrib[i].src_index] * 4;
234            assert(tab[vinfo->attrib[i].src_index] != -1);
235            switch (vinfo->attrib[i].interp_mode) {
236                case INTERP_LINEAR:
237                    rs->ip[col_count] |=
238                        R300_RS_COL_PTR(memory_pos) |
239                        R300_RS_COL_FMT(R300_RS_COL_FMT_RGBA);
240                    col_count++;
241                    break;
242                case INTERP_PERSPECTIVE:
243                    rs->ip[tex_count] |=
244                        R300_RS_TEX_PTR(memory_pos) |
245                        R300_RS_SEL_S(R300_RS_SEL_C0) |
246                        R300_RS_SEL_T(R300_RS_SEL_C1) |
247                        R300_RS_SEL_R(R300_RS_SEL_C2) |
248                        R300_RS_SEL_Q(R300_RS_SEL_C3);
249                    tex_count++;
250                    break;
251                default:
252                    break;
253            }
254        }
255
256        if (tex_count == 0) {
257            rs->ip[0] |=
258                R300_RS_SEL_S(R300_RS_SEL_K0) |
259                R300_RS_SEL_T(R300_RS_SEL_K0) |
260                R300_RS_SEL_R(R300_RS_SEL_K0) |
261                R300_RS_SEL_Q(R300_RS_SEL_K1);
262        }
263
264        for (i = 0; i < tex_count; i++) {
265            rs->inst[i] |= R300_RS_INST_TEX_ID(i) | R300_RS_INST_TEX_CN_WRITE |
266                R300_RS_INST_TEX_ADDR(fp_offset);
267            fp_offset++;
268        }
269
270        for (i = 0; i < col_count; i++) {
271            rs->inst[i] |= R300_RS_INST_COL_ID(i) | R300_RS_INST_COL_CN_WRITE |
272                R300_RS_INST_COL_ADDR(fp_offset);
273            fp_offset++;
274        }
275    }
276
277    rs->count = (tex_count * 4) | (col_count << R300_IC_COUNT_SHIFT) |
278        R300_HIRES_EN;
279
280    rs->inst_count = MAX2(MAX2(col_count - 1, tex_count - 1), 0);
281}
282
283void r300_update_derived_state(struct r300_context* r300)
284{
285    if (r300->dirty_state & R300_NEW_FRAGMENT_SHADER) {
286        r300_update_vertex_layout(r300);
287    }
288
289    if (r300->dirty_state & R300_NEW_VERTEX_FORMAT) {
290        r300_update_rs_block(r300);
291    }
292}
293