• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (C) 2014 Rob Clark <robclark@freedesktop.org>
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  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21  * SOFTWARE.
22  *
23  * Authors:
24  *    Rob Clark <robclark@freedesktop.org>
25  */
26 
27 #include "pipe/p_state.h"
28 #include "util/format/u_format.h"
29 #include "util/u_inlines.h"
30 #include "util/u_memory.h"
31 #include "util/u_string.h"
32 
33 #include "freedreno_program.h"
34 
35 #include "fd4_emit.h"
36 #include "fd4_format.h"
37 #include "fd4_program.h"
38 #include "fd4_texture.h"
39 
40 void
fd4_emit_shader(struct fd_ringbuffer * ring,const struct ir3_shader_variant * so)41 fd4_emit_shader(struct fd_ringbuffer *ring, const struct ir3_shader_variant *so)
42 {
43    const struct ir3_info *si = &so->info;
44    enum a4xx_state_block sb = fd4_stage2shadersb(so->type);
45    enum a4xx_state_src src;
46    uint32_t i, sz, *bin;
47 
48    if (FD_DBG(DIRECT)) {
49       sz = si->sizedwords;
50       src = SS4_DIRECT;
51       bin = fd_bo_map(so->bo);
52    } else {
53       sz = 0;
54       src = SS4_INDIRECT;
55       bin = NULL;
56    }
57 
58    OUT_PKT3(ring, CP_LOAD_STATE4, 2 + sz);
59    OUT_RING(ring, CP_LOAD_STATE4_0_DST_OFF(0) |
60                      CP_LOAD_STATE4_0_STATE_SRC(src) |
61                      CP_LOAD_STATE4_0_STATE_BLOCK(sb) |
62                      CP_LOAD_STATE4_0_NUM_UNIT(so->instrlen));
63    if (bin) {
64       OUT_RING(ring, CP_LOAD_STATE4_1_EXT_SRC_ADDR(0) |
65                         CP_LOAD_STATE4_1_STATE_TYPE(ST4_SHADER));
66    } else {
67       OUT_RELOC(ring, so->bo, 0, CP_LOAD_STATE4_1_STATE_TYPE(ST4_SHADER), 0);
68    }
69 
70    /* for how clever coverity is, it is sometimes rather dull, and
71     * doesn't realize that the only case where bin==NULL, sz==0:
72     */
73    assume(bin || (sz == 0));
74 
75    for (i = 0; i < sz; i++) {
76       OUT_RING(ring, bin[i]);
77    }
78 }
79 
80 struct stage {
81    const struct ir3_shader_variant *v;
82    const struct ir3_info *i;
83    /* const sizes are in units of 4 * vec4 */
84    uint8_t constoff;
85    uint8_t constlen;
86    /* instr sizes are in units of 16 instructions */
87    uint8_t instroff;
88    uint8_t instrlen;
89 };
90 
91 enum { VS = 0, FS = 1, HS = 2, DS = 3, GS = 4, MAX_STAGES };
92 
93 static void
setup_stages(struct fd4_emit * emit,struct stage * s)94 setup_stages(struct fd4_emit *emit, struct stage *s)
95 {
96    unsigned i;
97 
98    s[VS].v = fd4_emit_get_vp(emit);
99    s[FS].v = fd4_emit_get_fp(emit);
100 
101    s[HS].v = s[DS].v = s[GS].v = NULL; /* for now */
102 
103    for (i = 0; i < MAX_STAGES; i++) {
104       if (s[i].v) {
105          s[i].i = &s[i].v->info;
106          /* constlen is in units of 4 * vec4: */
107          assert(s[i].v->constlen % 4 == 0);
108          s[i].constlen = s[i].v->constlen / 4;
109          /* instrlen is already in units of 16 instr.. although
110           * probably we should ditch that and not make the compiler
111           * care about instruction group size of a3xx vs a4xx
112           */
113          s[i].instrlen = s[i].v->instrlen;
114       } else {
115          s[i].i = NULL;
116          s[i].constlen = 0;
117          s[i].instrlen = 0;
118       }
119    }
120 
121    /* NOTE: at least for gles2, blob partitions VS at bottom of const
122     * space and FS taking entire remaining space.  We probably don't
123     * need to do that the same way, but for now mimic what the blob
124     * does to make it easier to diff against register values from blob
125     *
126     * NOTE: if VS.instrlen + FS.instrlen > 64, then one or both shaders
127     * is run from external memory.
128     */
129    if ((s[VS].instrlen + s[FS].instrlen) > 64) {
130       /* prioritize FS for internal memory: */
131       if (s[FS].instrlen < 64) {
132          /* if FS can fit, kick VS out to external memory: */
133          s[VS].instrlen = 0;
134       } else if (s[VS].instrlen < 64) {
135          /* otherwise if VS can fit, kick out FS: */
136          s[FS].instrlen = 0;
137       } else {
138          /* neither can fit, run both from external memory: */
139          s[VS].instrlen = 0;
140          s[FS].instrlen = 0;
141       }
142    }
143    s[VS].constlen = 66;
144    s[FS].constlen = 128 - s[VS].constlen;
145    s[VS].instroff = 0;
146    s[VS].constoff = 0;
147    s[FS].instroff = 64 - s[FS].instrlen;
148    s[FS].constoff = s[VS].constlen;
149    s[HS].instroff = s[DS].instroff = s[GS].instroff = s[FS].instroff;
150    s[HS].constoff = s[DS].constoff = s[GS].constoff = s[FS].constoff;
151 }
152 
153 void
fd4_program_emit(struct fd_ringbuffer * ring,struct fd4_emit * emit,int nr,struct pipe_surface ** bufs)154 fd4_program_emit(struct fd_ringbuffer *ring, struct fd4_emit *emit, int nr,
155                  struct pipe_surface **bufs)
156 {
157    struct stage s[MAX_STAGES];
158    uint32_t pos_regid, posz_regid, psize_regid, color_regid[8];
159    uint32_t face_regid, coord_regid, zwcoord_regid, samp_id_regid,
160       samp_mask_regid, ij_regid[IJ_COUNT];
161    enum a3xx_threadsize fssz;
162    int constmode;
163    int i, j;
164 
165    assert(nr <= ARRAY_SIZE(color_regid));
166 
167    if (emit->binning_pass)
168       nr = 0;
169 
170    setup_stages(emit, s);
171 
172    fssz = (s[FS].i->double_threadsize) ? FOUR_QUADS : TWO_QUADS;
173 
174    /* blob seems to always use constmode currently: */
175    constmode = 1;
176 
177    pos_regid = ir3_find_output_regid(s[VS].v, VARYING_SLOT_POS);
178    if (pos_regid == regid(63, 0)) {
179       /* hw dislikes when there is no position output, which can
180        * happen for transform-feedback vertex shaders.  Just tell
181        * the hw to use r0.x, with whatever random value is there:
182        */
183       pos_regid = regid(0, 0);
184    }
185    posz_regid = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DEPTH);
186    psize_regid = ir3_find_output_regid(s[VS].v, VARYING_SLOT_PSIZ);
187    if (s[FS].v->color0_mrt) {
188       color_regid[0] = color_regid[1] = color_regid[2] = color_regid[3] =
189          color_regid[4] = color_regid[5] = color_regid[6] = color_regid[7] =
190             ir3_find_output_regid(s[FS].v, FRAG_RESULT_COLOR);
191    } else {
192       color_regid[0] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA0);
193       color_regid[1] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA1);
194       color_regid[2] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA2);
195       color_regid[3] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA3);
196       color_regid[4] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA4);
197       color_regid[5] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA5);
198       color_regid[6] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA6);
199       color_regid[7] = ir3_find_output_regid(s[FS].v, FRAG_RESULT_DATA7);
200    }
201 
202    samp_id_regid = ir3_find_sysval_regid(s[FS].v, SYSTEM_VALUE_SAMPLE_ID);
203    samp_mask_regid =
204       ir3_find_sysval_regid(s[FS].v, SYSTEM_VALUE_SAMPLE_MASK_IN);
205    face_regid = ir3_find_sysval_regid(s[FS].v, SYSTEM_VALUE_FRONT_FACE);
206    coord_regid = ir3_find_sysval_regid(s[FS].v, SYSTEM_VALUE_FRAG_COORD);
207    zwcoord_regid =
208       (coord_regid == regid(63, 0)) ? regid(63, 0) : (coord_regid + 2);
209    for (unsigned i = 0; i < ARRAY_SIZE(ij_regid); i++)
210       ij_regid[i] = ir3_find_sysval_regid(
211          s[FS].v, SYSTEM_VALUE_BARYCENTRIC_PERSP_PIXEL + i);
212 
213    /* we could probably divide this up into things that need to be
214     * emitted if frag-prog is dirty vs if vert-prog is dirty..
215     */
216 
217    OUT_PKT0(ring, REG_A4XX_HLSQ_UPDATE_CONTROL, 1);
218    OUT_RING(ring, 0x00000003);
219 
220    OUT_PKT0(ring, REG_A4XX_HLSQ_CONTROL_0_REG, 5);
221    OUT_RING(ring, A4XX_HLSQ_CONTROL_0_REG_FSTHREADSIZE(fssz) |
222                      A4XX_HLSQ_CONTROL_0_REG_CONSTMODE(constmode) |
223                      A4XX_HLSQ_CONTROL_0_REG_FSSUPERTHREADENABLE |
224                      /* NOTE:  I guess SHADERRESTART and CONSTFULLUPDATE maybe
225                       * flush some caches? I think we only need to set those
226                       * bits if we have updated const or shader..
227                       */
228                      A4XX_HLSQ_CONTROL_0_REG_SPSHADERRESTART |
229                      A4XX_HLSQ_CONTROL_0_REG_SPCONSTFULLUPDATE);
230    OUT_RING(ring, A4XX_HLSQ_CONTROL_1_REG_VSTHREADSIZE(TWO_QUADS) |
231                      A4XX_HLSQ_CONTROL_1_REG_VSSUPERTHREADENABLE |
232                      A4XX_HLSQ_CONTROL_1_REG_COORDREGID(coord_regid) |
233                      A4XX_HLSQ_CONTROL_1_REG_ZWCOORDREGID(zwcoord_regid));
234    OUT_RING(ring, A4XX_HLSQ_CONTROL_2_REG_PRIMALLOCTHRESHOLD(63) |
235                      A4XX_HLSQ_CONTROL_2_REG_SAMPLEID_REGID(samp_id_regid) |
236                      A4XX_HLSQ_CONTROL_2_REG_SAMPLEMASK_REGID(samp_mask_regid) |
237                      A4XX_HLSQ_CONTROL_2_REG_FACEREGID(face_regid));
238    /* XXX left out centroid/sample for now */
239    OUT_RING(
240       ring,
241       A4XX_HLSQ_CONTROL_3_REG_IJ_PERSP_PIXEL(ij_regid[IJ_PERSP_PIXEL]) |
242          A4XX_HLSQ_CONTROL_3_REG_IJ_LINEAR_PIXEL(ij_regid[IJ_LINEAR_PIXEL]) |
243          A4XX_HLSQ_CONTROL_3_REG_IJ_PERSP_CENTROID(
244             ij_regid[IJ_PERSP_CENTROID]) |
245          A4XX_HLSQ_CONTROL_3_REG_IJ_LINEAR_CENTROID(
246             ij_regid[IJ_LINEAR_CENTROID]));
247    OUT_RING(ring, 0x00fcfcfc); /* XXX HLSQ_CONTROL_4 */
248 
249    OUT_PKT0(ring, REG_A4XX_HLSQ_VS_CONTROL_REG, 5);
250    OUT_RING(ring,
251             A4XX_HLSQ_VS_CONTROL_REG_CONSTLENGTH(s[VS].constlen) |
252                A4XX_HLSQ_VS_CONTROL_REG_CONSTOBJECTOFFSET(s[VS].constoff) |
253                COND(s[VS].v && s[VS].v->has_ssbo, A4XX_HLSQ_VS_CONTROL_REG_SSBO_ENABLE) |
254                COND(s[VS].v, A4XX_HLSQ_VS_CONTROL_REG_ENABLED) |
255                A4XX_HLSQ_VS_CONTROL_REG_INSTRLENGTH(s[VS].instrlen) |
256                A4XX_HLSQ_VS_CONTROL_REG_SHADEROBJOFFSET(s[VS].instroff));
257    OUT_RING(ring,
258             A4XX_HLSQ_FS_CONTROL_REG_CONSTLENGTH(s[FS].constlen) |
259                A4XX_HLSQ_FS_CONTROL_REG_CONSTOBJECTOFFSET(s[FS].constoff) |
260                COND(s[FS].v && s[FS].v->has_ssbo, A4XX_HLSQ_FS_CONTROL_REG_SSBO_ENABLE) |
261                COND(s[FS].v, A4XX_HLSQ_FS_CONTROL_REG_ENABLED) |
262                A4XX_HLSQ_FS_CONTROL_REG_INSTRLENGTH(s[FS].instrlen) |
263                A4XX_HLSQ_FS_CONTROL_REG_SHADEROBJOFFSET(s[FS].instroff));
264    OUT_RING(ring,
265             A4XX_HLSQ_HS_CONTROL_REG_CONSTLENGTH(s[HS].constlen) |
266                A4XX_HLSQ_HS_CONTROL_REG_CONSTOBJECTOFFSET(s[HS].constoff) |
267                COND(s[HS].v && s[HS].v->has_ssbo, A4XX_HLSQ_HS_CONTROL_REG_SSBO_ENABLE) |
268                A4XX_HLSQ_HS_CONTROL_REG_INSTRLENGTH(s[HS].instrlen) |
269                A4XX_HLSQ_HS_CONTROL_REG_SHADEROBJOFFSET(s[HS].instroff));
270    OUT_RING(ring,
271             A4XX_HLSQ_DS_CONTROL_REG_CONSTLENGTH(s[DS].constlen) |
272                A4XX_HLSQ_DS_CONTROL_REG_CONSTOBJECTOFFSET(s[DS].constoff) |
273                COND(s[DS].v && s[DS].v->has_ssbo, A4XX_HLSQ_DS_CONTROL_REG_SSBO_ENABLE) |
274                A4XX_HLSQ_DS_CONTROL_REG_INSTRLENGTH(s[DS].instrlen) |
275                A4XX_HLSQ_DS_CONTROL_REG_SHADEROBJOFFSET(s[DS].instroff));
276    OUT_RING(ring,
277             A4XX_HLSQ_GS_CONTROL_REG_CONSTLENGTH(s[GS].constlen) |
278                A4XX_HLSQ_GS_CONTROL_REG_CONSTOBJECTOFFSET(s[GS].constoff) |
279                COND(s[GS].v && s[GS].v->has_ssbo, A4XX_HLSQ_GS_CONTROL_REG_SSBO_ENABLE) |
280                A4XX_HLSQ_GS_CONTROL_REG_INSTRLENGTH(s[GS].instrlen) |
281                A4XX_HLSQ_GS_CONTROL_REG_SHADEROBJOFFSET(s[GS].instroff));
282 
283    OUT_PKT0(ring, REG_A4XX_SP_SP_CTRL_REG, 1);
284    OUT_RING(ring,
285             0x140010 | /* XXX */
286                COND(emit->binning_pass, A4XX_SP_SP_CTRL_REG_BINNING_PASS));
287 
288    OUT_PKT0(ring, REG_A4XX_SP_INSTR_CACHE_CTRL, 1);
289    OUT_RING(ring, 0x7f | /* XXX */
290                      COND(s[VS].instrlen, A4XX_SP_INSTR_CACHE_CTRL_VS_BUFFER) |
291                      COND(s[FS].instrlen, A4XX_SP_INSTR_CACHE_CTRL_FS_BUFFER) |
292                      COND(s[VS].instrlen && s[FS].instrlen,
293                           A4XX_SP_INSTR_CACHE_CTRL_INSTR_BUFFER));
294 
295    OUT_PKT0(ring, REG_A4XX_SP_VS_LENGTH_REG, 1);
296    OUT_RING(ring, s[VS].v->instrlen); /* SP_VS_LENGTH_REG */
297 
298    OUT_PKT0(ring, REG_A4XX_SP_VS_CTRL_REG0, 3);
299    OUT_RING(
300       ring,
301       A4XX_SP_VS_CTRL_REG0_THREADMODE(MULTI) |
302          A4XX_SP_VS_CTRL_REG0_HALFREGFOOTPRINT(s[VS].i->max_half_reg + 1) |
303          A4XX_SP_VS_CTRL_REG0_FULLREGFOOTPRINT(s[VS].i->max_reg + 1) |
304          A4XX_SP_VS_CTRL_REG0_INOUTREGOVERLAP(0) |
305          A4XX_SP_VS_CTRL_REG0_THREADSIZE(TWO_QUADS) |
306          A4XX_SP_VS_CTRL_REG0_SUPERTHREADMODE |
307          COND(s[VS].v->need_pixlod, A4XX_SP_VS_CTRL_REG0_PIXLODENABLE));
308    OUT_RING(ring,
309             A4XX_SP_VS_CTRL_REG1_CONSTLENGTH(s[VS].constlen) |
310                A4XX_SP_VS_CTRL_REG1_INITIALOUTSTANDING(s[VS].v->total_in));
311    OUT_RING(ring, A4XX_SP_VS_PARAM_REG_POSREGID(pos_regid) |
312                      A4XX_SP_VS_PARAM_REG_PSIZEREGID(psize_regid) |
313                      A4XX_SP_VS_PARAM_REG_TOTALVSOUTVAR(s[FS].v->varying_in));
314 
315    struct ir3_shader_linkage l = {0};
316    ir3_link_shaders(&l, s[VS].v, s[FS].v, false);
317 
318    for (i = 0, j = 0; (i < 16) && (j < l.cnt); i++) {
319       uint32_t reg = 0;
320 
321       OUT_PKT0(ring, REG_A4XX_SP_VS_OUT_REG(i), 1);
322 
323       reg |= A4XX_SP_VS_OUT_REG_A_REGID(l.var[j].regid);
324       reg |= A4XX_SP_VS_OUT_REG_A_COMPMASK(l.var[j].compmask);
325       j++;
326 
327       reg |= A4XX_SP_VS_OUT_REG_B_REGID(l.var[j].regid);
328       reg |= A4XX_SP_VS_OUT_REG_B_COMPMASK(l.var[j].compmask);
329       j++;
330 
331       OUT_RING(ring, reg);
332    }
333 
334    for (i = 0, j = 0; (i < 8) && (j < l.cnt); i++) {
335       uint32_t reg = 0;
336 
337       OUT_PKT0(ring, REG_A4XX_SP_VS_VPC_DST_REG(i), 1);
338 
339       reg |= A4XX_SP_VS_VPC_DST_REG_OUTLOC0(l.var[j++].loc + 8);
340       reg |= A4XX_SP_VS_VPC_DST_REG_OUTLOC1(l.var[j++].loc + 8);
341       reg |= A4XX_SP_VS_VPC_DST_REG_OUTLOC2(l.var[j++].loc + 8);
342       reg |= A4XX_SP_VS_VPC_DST_REG_OUTLOC3(l.var[j++].loc + 8);
343 
344       OUT_RING(ring, reg);
345    }
346 
347    OUT_PKT0(ring, REG_A4XX_SP_VS_OBJ_OFFSET_REG, 2);
348    OUT_RING(ring, A4XX_SP_VS_OBJ_OFFSET_REG_CONSTOBJECTOFFSET(s[VS].constoff) |
349                      A4XX_SP_VS_OBJ_OFFSET_REG_SHADEROBJOFFSET(s[VS].instroff));
350    OUT_RELOC(ring, s[VS].v->bo, 0, 0, 0); /* SP_VS_OBJ_START_REG */
351 
352    if (emit->binning_pass) {
353       OUT_PKT0(ring, REG_A4XX_SP_FS_LENGTH_REG, 1);
354       OUT_RING(ring, 0x00000000); /* SP_FS_LENGTH_REG */
355 
356       OUT_PKT0(ring, REG_A4XX_SP_FS_CTRL_REG0, 2);
357       OUT_RING(ring,
358                A4XX_SP_FS_CTRL_REG0_THREADMODE(MULTI) |
359                   COND(s[FS].v->total_in > 0, A4XX_SP_FS_CTRL_REG0_VARYING) |
360                   A4XX_SP_FS_CTRL_REG0_HALFREGFOOTPRINT(0) |
361                   A4XX_SP_FS_CTRL_REG0_FULLREGFOOTPRINT(0) |
362                   A4XX_SP_FS_CTRL_REG0_INOUTREGOVERLAP(1) |
363                   A4XX_SP_FS_CTRL_REG0_THREADSIZE(fssz) |
364                   A4XX_SP_FS_CTRL_REG0_SUPERTHREADMODE);
365       OUT_RING(ring,
366                A4XX_SP_FS_CTRL_REG1_CONSTLENGTH(s[FS].constlen) | 0x80000000);
367 
368       OUT_PKT0(ring, REG_A4XX_SP_FS_OBJ_OFFSET_REG, 2);
369       OUT_RING(ring,
370                A4XX_SP_FS_OBJ_OFFSET_REG_CONSTOBJECTOFFSET(s[FS].constoff) |
371                   A4XX_SP_FS_OBJ_OFFSET_REG_SHADEROBJOFFSET(s[FS].instroff));
372       OUT_RING(ring, 0x00000000);
373    } else {
374       OUT_PKT0(ring, REG_A4XX_SP_FS_LENGTH_REG, 1);
375       OUT_RING(ring, s[FS].v->instrlen); /* SP_FS_LENGTH_REG */
376 
377       OUT_PKT0(ring, REG_A4XX_SP_FS_CTRL_REG0, 2);
378       OUT_RING(
379          ring,
380          A4XX_SP_FS_CTRL_REG0_THREADMODE(MULTI) |
381             COND(s[FS].v->total_in > 0, A4XX_SP_FS_CTRL_REG0_VARYING) |
382             A4XX_SP_FS_CTRL_REG0_HALFREGFOOTPRINT(s[FS].i->max_half_reg + 1) |
383             A4XX_SP_FS_CTRL_REG0_FULLREGFOOTPRINT(s[FS].i->max_reg + 1) |
384             A4XX_SP_FS_CTRL_REG0_INOUTREGOVERLAP(1) |
385             A4XX_SP_FS_CTRL_REG0_THREADSIZE(fssz) |
386             A4XX_SP_FS_CTRL_REG0_SUPERTHREADMODE |
387             COND(s[FS].v->need_pixlod, A4XX_SP_FS_CTRL_REG0_PIXLODENABLE));
388       OUT_RING(ring,
389                A4XX_SP_FS_CTRL_REG1_CONSTLENGTH(s[FS].constlen) |
390                   0x80000000 | /* XXX */
391                   COND(s[FS].v->frag_face, A4XX_SP_FS_CTRL_REG1_FACENESS) |
392                   COND(s[FS].v->total_in > 0, A4XX_SP_FS_CTRL_REG1_VARYING) |
393                   COND(s[FS].v->fragcoord_compmask != 0,
394                        A4XX_SP_FS_CTRL_REG1_FRAGCOORD));
395 
396       OUT_PKT0(ring, REG_A4XX_SP_FS_OBJ_OFFSET_REG, 2);
397       OUT_RING(ring,
398                A4XX_SP_FS_OBJ_OFFSET_REG_CONSTOBJECTOFFSET(s[FS].constoff) |
399                   A4XX_SP_FS_OBJ_OFFSET_REG_SHADEROBJOFFSET(s[FS].instroff));
400       OUT_RELOC(ring, s[FS].v->bo, 0, 0, 0); /* SP_FS_OBJ_START_REG */
401    }
402 
403    OUT_PKT0(ring, REG_A4XX_SP_HS_OBJ_OFFSET_REG, 1);
404    OUT_RING(ring, A4XX_SP_HS_OBJ_OFFSET_REG_CONSTOBJECTOFFSET(s[HS].constoff) |
405                      A4XX_SP_HS_OBJ_OFFSET_REG_SHADEROBJOFFSET(s[HS].instroff));
406 
407    OUT_PKT0(ring, REG_A4XX_SP_DS_OBJ_OFFSET_REG, 1);
408    OUT_RING(ring, A4XX_SP_DS_OBJ_OFFSET_REG_CONSTOBJECTOFFSET(s[DS].constoff) |
409                      A4XX_SP_DS_OBJ_OFFSET_REG_SHADEROBJOFFSET(s[DS].instroff));
410 
411    OUT_PKT0(ring, REG_A4XX_SP_GS_OBJ_OFFSET_REG, 1);
412    OUT_RING(ring, A4XX_SP_GS_OBJ_OFFSET_REG_CONSTOBJECTOFFSET(s[GS].constoff) |
413                      A4XX_SP_GS_OBJ_OFFSET_REG_SHADEROBJOFFSET(s[GS].instroff));
414 
415    OUT_PKT0(ring, REG_A4XX_GRAS_CNTL, 1);
416    OUT_RING(ring,
417             CONDREG(face_regid, A4XX_GRAS_CNTL_IJ_PERSP) |
418                CONDREG(zwcoord_regid, A4XX_GRAS_CNTL_IJ_PERSP) |
419                CONDREG(ij_regid[IJ_PERSP_PIXEL], A4XX_GRAS_CNTL_IJ_PERSP) |
420                CONDREG(ij_regid[IJ_LINEAR_PIXEL], A4XX_GRAS_CNTL_IJ_LINEAR) |
421                CONDREG(ij_regid[IJ_PERSP_CENTROID], A4XX_GRAS_CNTL_IJ_PERSP));
422 
423    OUT_PKT0(ring, REG_A4XX_RB_RENDER_CONTROL2, 1);
424    OUT_RING(
425       ring,
426       A4XX_RB_RENDER_CONTROL2_MSAA_SAMPLES(0) |
427          CONDREG(ij_regid[IJ_PERSP_PIXEL],
428                  A4XX_RB_RENDER_CONTROL2_IJ_PERSP_PIXEL) |
429          CONDREG(ij_regid[IJ_PERSP_CENTROID],
430                  A4XX_RB_RENDER_CONTROL2_IJ_PERSP_CENTROID) |
431          CONDREG(ij_regid[IJ_LINEAR_PIXEL], A4XX_RB_RENDER_CONTROL2_SIZE) |
432          CONDREG(samp_id_regid, A4XX_RB_RENDER_CONTROL2_SAMPLEID) |
433          COND(s[FS].v->frag_face, A4XX_RB_RENDER_CONTROL2_FACENESS) |
434          CONDREG(samp_mask_regid, A4XX_RB_RENDER_CONTROL2_SAMPLEMASK) |
435          COND(s[FS].v->fragcoord_compmask != 0,
436               A4XX_RB_RENDER_CONTROL2_COORD_MASK(s[FS].v->fragcoord_compmask)));
437 
438    OUT_PKT0(ring, REG_A4XX_RB_FS_OUTPUT_REG, 1);
439    OUT_RING(ring,
440             A4XX_RB_FS_OUTPUT_REG_MRT(nr) |
441                COND(s[FS].v->writes_pos, A4XX_RB_FS_OUTPUT_REG_FRAG_WRITES_Z));
442 
443    OUT_PKT0(ring, REG_A4XX_SP_FS_OUTPUT_REG, 1);
444    OUT_RING(ring,
445             A4XX_SP_FS_OUTPUT_REG_MRT(nr) |
446                COND(s[FS].v->writes_pos, A4XX_SP_FS_OUTPUT_REG_DEPTH_ENABLE) |
447                A4XX_SP_FS_OUTPUT_REG_DEPTH_REGID(posz_regid));
448 
449    OUT_PKT0(ring, REG_A4XX_SP_FS_MRT_REG(0), 8);
450    for (i = 0; i < 8; i++) {
451       enum a4xx_color_fmt format = 0;
452       bool srgb = false;
453       bool uint = false;
454       bool sint = false;
455       if (i < nr) {
456          format = fd4_emit_format(bufs[i]);
457          if (bufs[i]) {
458             if (!emit->no_decode_srgb)
459                srgb = util_format_is_srgb(bufs[i]->format);
460             uint = util_format_is_pure_uint(bufs[i]->format);
461             sint = util_format_is_pure_sint(bufs[i]->format);
462          }
463       }
464       OUT_RING(ring, A4XX_SP_FS_MRT_REG_REGID(color_regid[i]) |
465                         A4XX_SP_FS_MRT_REG_MRTFORMAT(format) |
466                         COND(srgb, A4XX_SP_FS_MRT_REG_COLOR_SRGB) |
467                         COND(uint, A4XX_SP_FS_MRT_REG_COLOR_UINT) |
468                         COND(sint, A4XX_SP_FS_MRT_REG_COLOR_SINT) |
469                         COND(color_regid[i] & HALF_REG_ID,
470                              A4XX_SP_FS_MRT_REG_HALF_PRECISION));
471    }
472 
473    if (emit->binning_pass) {
474       OUT_PKT0(ring, REG_A4XX_VPC_ATTR, 2);
475       OUT_RING(ring, A4XX_VPC_ATTR_THRDASSIGN(1) | 0x40000000 | /* XXX */
476                         COND(s[VS].v->writes_psize, A4XX_VPC_ATTR_PSIZE));
477       OUT_RING(ring, 0x00000000);
478    } else {
479       uint32_t vinterp[8], vpsrepl[8];
480 
481       memset(vinterp, 0, sizeof(vinterp));
482       memset(vpsrepl, 0, sizeof(vpsrepl));
483 
484       /* looks like we need to do int varyings in the frag
485        * shader on a4xx (no flatshad reg?  or a420.0 bug?):
486        *
487        *    (sy)(ss)nop
488        *    (sy)ldlv.u32 r0.x,l[r0.x], 1
489        *    ldlv.u32 r0.y,l[r0.x+1], 1
490        *    (ss)bary.f (ei)r63.x, 0, r0.x
491        *    (ss)(rpt1)cov.s32f16 hr0.x, (r)r0.x
492        *    (rpt5)nop
493        *    sam (f16)(xyzw)hr0.x, hr0.x, s#0, t#0
494        *
495        * Possibly on later a4xx variants we'll be able to use
496        * something like the code below instead of workaround
497        * in the shader:
498        */
499       /* figure out VARYING_INTERP / VARYING_PS_REPL register values: */
500       for (j = -1;
501            (j = ir3_next_varying(s[FS].v, j)) < (int)s[FS].v->inputs_count;) {
502          /* NOTE: varyings are packed, so if compmask is 0xb
503           * then first, third, and fourth component occupy
504           * three consecutive varying slots:
505           */
506          unsigned compmask = s[FS].v->inputs[j].compmask;
507 
508          uint32_t inloc = s[FS].v->inputs[j].inloc;
509 
510          if (s[FS].v->inputs[j].flat ||
511              (s[FS].v->inputs[j].rasterflat && emit->rasterflat)) {
512             uint32_t loc = inloc;
513 
514             for (i = 0; i < 4; i++) {
515                if (compmask & (1 << i)) {
516                   vinterp[loc / 16] |= 1 << ((loc % 16) * 2);
517                   // flatshade[loc / 32] |= 1 << (loc % 32);
518                   loc++;
519                }
520             }
521          }
522 
523          bool coord_mode = emit->sprite_coord_mode;
524          if (ir3_point_sprite(s[FS].v, j, emit->sprite_coord_enable,
525                               &coord_mode)) {
526             /* mask is two 2-bit fields, where:
527              *   '01' -> S
528              *   '10' -> T
529              *   '11' -> 1 - T  (flip mode)
530              */
531             unsigned mask = coord_mode ? 0b1101 : 0b1001;
532             uint32_t loc = inloc;
533             if (compmask & 0x1) {
534                vpsrepl[loc / 16] |= ((mask >> 0) & 0x3) << ((loc % 16) * 2);
535                loc++;
536             }
537             if (compmask & 0x2) {
538                vpsrepl[loc / 16] |= ((mask >> 2) & 0x3) << ((loc % 16) * 2);
539                loc++;
540             }
541             if (compmask & 0x4) {
542                /* .z <- 0.0f */
543                vinterp[loc / 16] |= 0b10 << ((loc % 16) * 2);
544                loc++;
545             }
546             if (compmask & 0x8) {
547                /* .w <- 1.0f */
548                vinterp[loc / 16] |= 0b11 << ((loc % 16) * 2);
549                loc++;
550             }
551          }
552       }
553 
554       OUT_PKT0(ring, REG_A4XX_VPC_ATTR, 2);
555       OUT_RING(ring, A4XX_VPC_ATTR_TOTALATTR(s[FS].v->total_in) |
556                         A4XX_VPC_ATTR_THRDASSIGN(1) |
557                         COND(s[FS].v->total_in > 0, A4XX_VPC_ATTR_ENABLE) |
558                         0x40000000 | /* XXX */
559                         COND(s[VS].v->writes_psize, A4XX_VPC_ATTR_PSIZE));
560       OUT_RING(ring, A4XX_VPC_PACK_NUMFPNONPOSVAR(s[FS].v->total_in) |
561                         A4XX_VPC_PACK_NUMNONPOSVSVAR(s[FS].v->total_in));
562 
563       OUT_PKT0(ring, REG_A4XX_VPC_VARYING_INTERP_MODE(0), 8);
564       for (i = 0; i < 8; i++)
565          OUT_RING(ring, vinterp[i]); /* VPC_VARYING_INTERP[i].MODE */
566 
567       OUT_PKT0(ring, REG_A4XX_VPC_VARYING_PS_REPL_MODE(0), 8);
568       for (i = 0; i < 8; i++)
569          OUT_RING(ring, vpsrepl[i]); /* VPC_VARYING_PS_REPL[i] */
570    }
571 
572    if (s[VS].instrlen)
573       fd4_emit_shader(ring, s[VS].v);
574 
575    if (!emit->binning_pass)
576       if (s[FS].instrlen)
577          fd4_emit_shader(ring, s[FS].v);
578 }
579 
580 static struct ir3_program_state *
fd4_program_create(void * data,struct ir3_shader_variant * bs,struct ir3_shader_variant * vs,struct ir3_shader_variant * hs,struct ir3_shader_variant * ds,struct ir3_shader_variant * gs,struct ir3_shader_variant * fs,const struct ir3_cache_key * key)581 fd4_program_create(void *data, struct ir3_shader_variant *bs,
582                    struct ir3_shader_variant *vs, struct ir3_shader_variant *hs,
583                    struct ir3_shader_variant *ds, struct ir3_shader_variant *gs,
584                    struct ir3_shader_variant *fs,
585                    const struct ir3_cache_key *key) in_dt
586 {
587    struct fd_context *ctx = fd_context(data);
588    struct fd4_program_state *state = CALLOC_STRUCT(fd4_program_state);
589 
590    tc_assert_driver_thread(ctx->tc);
591 
592    state->bs = bs;
593    state->vs = vs;
594    state->fs = fs;
595 
596    return &state->base;
597 }
598 
599 static void
fd4_program_destroy(void * data,struct ir3_program_state * state)600 fd4_program_destroy(void *data, struct ir3_program_state *state)
601 {
602    struct fd4_program_state *so = fd4_program_state(state);
603    free(so);
604 }
605 
606 static const struct ir3_cache_funcs cache_funcs = {
607    .create_state = fd4_program_create,
608    .destroy_state = fd4_program_destroy,
609 };
610 
611 void
fd4_prog_init(struct pipe_context * pctx)612 fd4_prog_init(struct pipe_context *pctx)
613 {
614    struct fd_context *ctx = fd_context(pctx);
615 
616    ctx->shader_cache = ir3_cache_create(&cache_funcs, ctx);
617    ir3_prog_init(pctx);
618    fd_prog_init(pctx);
619 }
620