1 /*
2 * Copyright © 2012 Rob Clark <robclark@freedesktop.org>
3 * SPDX-License-Identifier: MIT
4 *
5 * Authors:
6 * Rob Clark <robclark@freedesktop.org>
7 * Jonathan Marek <jonathan@marek.ca>
8 */
9
10 #include "nir/tgsi_to_nir.h"
11 #include "pipe/p_state.h"
12 #include "tgsi/tgsi_dump.h"
13 #include "util/format/u_format.h"
14 #include "util/u_inlines.h"
15 #include "util/u_memory.h"
16 #include "util/u_string.h"
17
18 #include "freedreno_program.h"
19
20 #include "ir2/instr-a2xx.h"
21 #include "fd2_program.h"
22 #include "fd2_texture.h"
23 #include "fd2_util.h"
24 #include "ir2.h"
25
26 static struct fd2_shader_stateobj *
create_shader(struct pipe_context * pctx,gl_shader_stage type)27 create_shader(struct pipe_context *pctx, gl_shader_stage type)
28 {
29 struct fd2_shader_stateobj *so = CALLOC_STRUCT(fd2_shader_stateobj);
30 if (!so)
31 return NULL;
32 so->type = type;
33 so->is_a20x = is_a20x(fd_context(pctx)->screen);
34 return so;
35 }
36
37 static void
delete_shader(struct fd2_shader_stateobj * so)38 delete_shader(struct fd2_shader_stateobj *so)
39 {
40 if (!so)
41 return;
42 ralloc_free(so->nir);
43 for (int i = 0; i < ARRAY_SIZE(so->variant); i++)
44 free(so->variant[i].info.dwords);
45 free(so);
46 }
47
48 static void
emit(struct fd_ringbuffer * ring,gl_shader_stage type,struct ir2_shader_info * info,struct util_dynarray * patches)49 emit(struct fd_ringbuffer *ring, gl_shader_stage type,
50 struct ir2_shader_info *info, struct util_dynarray *patches)
51 {
52 unsigned i;
53
54 assert(info->sizedwords);
55
56 OUT_PKT3(ring, CP_IM_LOAD_IMMEDIATE, 2 + info->sizedwords);
57 OUT_RING(ring, type == MESA_SHADER_FRAGMENT);
58 OUT_RING(ring, info->sizedwords);
59
60 if (patches)
61 util_dynarray_append(patches, uint32_t *,
62 &ring->cur[info->mem_export_ptr]);
63
64 for (i = 0; i < info->sizedwords; i++)
65 OUT_RING(ring, info->dwords[i]);
66 }
67
68 static int
ir2_glsl_type_size(const struct glsl_type * type,bool bindless)69 ir2_glsl_type_size(const struct glsl_type *type, bool bindless)
70 {
71 return glsl_count_attribute_slots(type, false);
72 }
73
74 static void *
fd2_fp_state_create(struct pipe_context * pctx,const struct pipe_shader_state * cso)75 fd2_fp_state_create(struct pipe_context *pctx,
76 const struct pipe_shader_state *cso)
77 {
78 struct fd2_shader_stateobj *so = create_shader(pctx, MESA_SHADER_FRAGMENT);
79 if (!so)
80 return NULL;
81
82 so->nir = (cso->type == PIPE_SHADER_IR_NIR)
83 ? cso->ir.nir
84 : tgsi_to_nir(cso->tokens, pctx->screen, false);
85
86 NIR_PASS_V(so->nir, nir_lower_io, nir_var_shader_in | nir_var_shader_out,
87 ir2_glsl_type_size,
88 nir_lower_io_use_interpolated_input_intrinsics);
89
90 if (ir2_optimize_nir(so->nir, true))
91 goto fail;
92
93 so->first_immediate = so->nir->num_uniforms;
94
95 ir2_compile(so, 0, NULL);
96
97 ralloc_free(so->nir);
98 so->nir = NULL;
99 return so;
100
101 fail:
102 delete_shader(so);
103 return NULL;
104 }
105
106 static void
fd2_fp_state_delete(struct pipe_context * pctx,void * hwcso)107 fd2_fp_state_delete(struct pipe_context *pctx, void *hwcso)
108 {
109 struct fd2_shader_stateobj *so = hwcso;
110 delete_shader(so);
111 }
112
113 static void *
fd2_vp_state_create(struct pipe_context * pctx,const struct pipe_shader_state * cso)114 fd2_vp_state_create(struct pipe_context *pctx,
115 const struct pipe_shader_state *cso)
116 {
117 struct fd2_shader_stateobj *so = create_shader(pctx, MESA_SHADER_VERTEX);
118 if (!so)
119 return NULL;
120
121 so->nir = (cso->type == PIPE_SHADER_IR_NIR)
122 ? cso->ir.nir
123 : tgsi_to_nir(cso->tokens, pctx->screen, false);
124
125 NIR_PASS_V(so->nir, nir_lower_io, nir_var_shader_in | nir_var_shader_out,
126 ir2_glsl_type_size, (nir_lower_io_options)0);
127
128 if (ir2_optimize_nir(so->nir, true))
129 goto fail;
130
131 so->first_immediate = so->nir->num_uniforms;
132
133 /* compile binning variant now */
134 ir2_compile(so, 0, NULL);
135
136 return so;
137
138 fail:
139 delete_shader(so);
140 return NULL;
141 }
142
143 static void
fd2_vp_state_delete(struct pipe_context * pctx,void * hwcso)144 fd2_vp_state_delete(struct pipe_context *pctx, void *hwcso)
145 {
146 struct fd2_shader_stateobj *so = hwcso;
147 delete_shader(so);
148 }
149
150 static void
patch_vtx_fetch(struct fd_context * ctx,struct pipe_vertex_element * elem,instr_fetch_vtx_t * instr,uint16_t dst_swiz)151 patch_vtx_fetch(struct fd_context *ctx, struct pipe_vertex_element *elem,
152 instr_fetch_vtx_t *instr, uint16_t dst_swiz) assert_dt
153 {
154 struct surface_format fmt = fd2_pipe2surface(elem->src_format);
155
156 instr->dst_swiz = fd2_vtx_swiz(elem->src_format, dst_swiz);
157 instr->format_comp_all = fmt.sign == SQ_TEX_SIGN_SIGNED;
158 instr->num_format_all = fmt.num_format;
159 instr->format = fmt.format;
160 instr->exp_adjust_all = fmt.exp_adjust;
161 instr->stride = elem->src_stride;
162 instr->offset = elem->src_offset;
163 }
164
165 static void
patch_fetches(struct fd_context * ctx,struct ir2_shader_info * info,struct fd_vertex_stateobj * vtx,struct fd_texture_stateobj * tex)166 patch_fetches(struct fd_context *ctx, struct ir2_shader_info *info,
167 struct fd_vertex_stateobj *vtx,
168 struct fd_texture_stateobj *tex) assert_dt
169 {
170 for (int i = 0; i < info->num_fetch_instrs; i++) {
171 struct ir2_fetch_info *fi = &info->fetch_info[i];
172
173 instr_fetch_t *instr = (instr_fetch_t *)&info->dwords[fi->offset];
174 if (instr->opc == VTX_FETCH) {
175 unsigned idx =
176 (instr->vtx.const_index - 20) * 3 + instr->vtx.const_index_sel;
177 patch_vtx_fetch(ctx, &vtx->pipe[idx], &instr->vtx, fi->vtx.dst_swiz);
178 continue;
179 }
180
181 assert(instr->opc == TEX_FETCH);
182 instr->tex.const_idx = fd2_get_const_idx(ctx, tex, fi->tex.samp_id);
183 instr->tex.src_swiz = fi->tex.src_swiz;
184 }
185 }
186
187 void
fd2_program_emit(struct fd_context * ctx,struct fd_ringbuffer * ring,struct fd_program_stateobj * prog)188 fd2_program_emit(struct fd_context *ctx, struct fd_ringbuffer *ring,
189 struct fd_program_stateobj *prog)
190 {
191 struct fd2_shader_stateobj *fp = NULL, *vp;
192 struct ir2_shader_info *fpi, *vpi;
193 struct ir2_frag_linkage *f;
194 uint8_t vs_gprs, fs_gprs = 0, vs_export = 0;
195 enum a2xx_sq_ps_vtx_mode mode = POSITION_1_VECTOR;
196 bool binning = (ctx->batch && ring == ctx->batch->binning);
197 unsigned variant = 0;
198
199 vp = prog->vs;
200
201 /* find variant matching the linked fragment shader */
202 if (!binning) {
203 fp = prog->fs;
204 for (variant = 1; variant < ARRAY_SIZE(vp->variant); variant++) {
205 /* if checked all variants, compile a new variant */
206 if (!vp->variant[variant].info.sizedwords) {
207 ir2_compile(vp, variant, fp);
208 break;
209 }
210
211 /* check if fragment shader linkage matches */
212 if (!memcmp(&vp->variant[variant].f, &fp->variant[0].f,
213 sizeof(struct ir2_frag_linkage)))
214 break;
215 }
216 assert(variant < ARRAY_SIZE(vp->variant));
217 }
218
219 vpi = &vp->variant[variant].info;
220 fpi = &fp->variant[0].info;
221 f = &fp->variant[0].f;
222
223 /* clear/gmem2mem/mem2gmem need to be changed to remove this condition */
224 if (prog != &ctx->solid_prog && prog != &ctx->blit_prog[0]) {
225 patch_fetches(ctx, vpi, ctx->vtx.vtx, &ctx->tex[PIPE_SHADER_VERTEX]);
226 if (fp)
227 patch_fetches(ctx, fpi, NULL, &ctx->tex[PIPE_SHADER_FRAGMENT]);
228 }
229
230 emit(ring, MESA_SHADER_VERTEX, vpi,
231 binning ? &ctx->batch->shader_patches : NULL);
232
233 if (fp) {
234 emit(ring, MESA_SHADER_FRAGMENT, fpi, NULL);
235 fs_gprs = (fpi->max_reg < 0) ? 0x80 : fpi->max_reg;
236 vs_export = MAX2(1, f->inputs_count) - 1;
237 }
238
239 vs_gprs = (vpi->max_reg < 0) ? 0x80 : vpi->max_reg;
240
241 if (vp->writes_psize && !binning)
242 mode = POSITION_2_VECTORS_SPRITE;
243
244 /* set register to use for param (fragcoord/pointcoord/frontfacing) */
245 OUT_PKT3(ring, CP_SET_CONSTANT, 2);
246 OUT_RING(ring, CP_REG(REG_A2XX_SQ_CONTEXT_MISC));
247 OUT_RING(ring,
248 A2XX_SQ_CONTEXT_MISC_SC_SAMPLE_CNTL(CENTERS_ONLY) |
249 COND(fp, A2XX_SQ_CONTEXT_MISC_PARAM_GEN_POS(f->inputs_count)) |
250 /* we need SCREEN_XY for both fragcoord and frontfacing */
251 A2XX_SQ_CONTEXT_MISC_SC_OUTPUT_SCREEN_XY);
252
253 OUT_PKT3(ring, CP_SET_CONSTANT, 2);
254 OUT_RING(ring, CP_REG(REG_A2XX_SQ_PROGRAM_CNTL));
255 OUT_RING(ring,
256 A2XX_SQ_PROGRAM_CNTL_PS_EXPORT_MODE(2) |
257 A2XX_SQ_PROGRAM_CNTL_VS_EXPORT_MODE(mode) |
258 A2XX_SQ_PROGRAM_CNTL_VS_RESOURCE |
259 A2XX_SQ_PROGRAM_CNTL_PS_RESOURCE |
260 A2XX_SQ_PROGRAM_CNTL_VS_EXPORT_COUNT(vs_export) |
261 A2XX_SQ_PROGRAM_CNTL_PS_REGS(fs_gprs) |
262 A2XX_SQ_PROGRAM_CNTL_VS_REGS(vs_gprs) |
263 COND(fp && fp->need_param, A2XX_SQ_PROGRAM_CNTL_PARAM_GEN) |
264 COND(!fp, A2XX_SQ_PROGRAM_CNTL_GEN_INDEX_VTX));
265 }
266
267 void
fd2_prog_init(struct pipe_context * pctx)268 fd2_prog_init(struct pipe_context *pctx)
269 {
270 struct fd_context *ctx = fd_context(pctx);
271 struct fd_program_stateobj *prog;
272 struct fd2_shader_stateobj *so;
273 struct ir2_shader_info *info;
274 instr_fetch_vtx_t *instr;
275
276 pctx->create_fs_state = fd2_fp_state_create;
277 pctx->delete_fs_state = fd2_fp_state_delete;
278
279 pctx->create_vs_state = fd2_vp_state_create;
280 pctx->delete_vs_state = fd2_vp_state_delete;
281
282 fd_prog_init(pctx);
283
284 /* XXX maybe its possible to reuse patch_vtx_fetch somehow? */
285
286 prog = &ctx->solid_prog;
287 so = prog->vs;
288 ir2_compile(prog->vs, 1, prog->fs);
289
290 #define IR2_FETCH_SWIZ_XY01 0xb08
291 #define IR2_FETCH_SWIZ_XYZ1 0xa88
292
293 info = &so->variant[1].info;
294
295 instr = (instr_fetch_vtx_t *)&info->dwords[info->fetch_info[0].offset];
296 instr->const_index = 26;
297 instr->const_index_sel = 0;
298 instr->format = FMT_32_32_32_FLOAT;
299 instr->format_comp_all = false;
300 instr->stride = 12;
301 instr->num_format_all = true;
302 instr->dst_swiz = IR2_FETCH_SWIZ_XYZ1;
303
304 prog = &ctx->blit_prog[0];
305 so = prog->vs;
306 ir2_compile(prog->vs, 1, prog->fs);
307
308 info = &so->variant[1].info;
309
310 instr = (instr_fetch_vtx_t *)&info->dwords[info->fetch_info[0].offset];
311 instr->const_index = 26;
312 instr->const_index_sel = 1;
313 instr->format = FMT_32_32_FLOAT;
314 instr->format_comp_all = false;
315 instr->stride = 8;
316 instr->num_format_all = false;
317 instr->dst_swiz = IR2_FETCH_SWIZ_XY01;
318
319 instr = (instr_fetch_vtx_t *)&info->dwords[info->fetch_info[1].offset];
320 instr->const_index = 26;
321 instr->const_index_sel = 0;
322 instr->format = FMT_32_32_32_FLOAT;
323 instr->format_comp_all = false;
324 instr->stride = 12;
325 instr->num_format_all = false;
326 instr->dst_swiz = IR2_FETCH_SWIZ_XYZ1;
327 }
328