• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (C) 2012 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  *    Jonathan Marek <jonathan@marek.ca>
26  */
27 
28 #include "nir/tgsi_to_nir.h"
29 #include "pipe/p_state.h"
30 #include "tgsi/tgsi_dump.h"
31 #include "util/format/u_format.h"
32 #include "util/u_inlines.h"
33 #include "util/u_memory.h"
34 #include "util/u_string.h"
35 
36 #include "freedreno_program.h"
37 
38 #include "ir2/instr-a2xx.h"
39 #include "fd2_program.h"
40 #include "fd2_texture.h"
41 #include "fd2_util.h"
42 #include "ir2.h"
43 
44 static struct fd2_shader_stateobj *
create_shader(struct pipe_context * pctx,gl_shader_stage type)45 create_shader(struct pipe_context *pctx, gl_shader_stage type)
46 {
47    struct fd2_shader_stateobj *so = CALLOC_STRUCT(fd2_shader_stateobj);
48    if (!so)
49       return NULL;
50    so->type = type;
51    so->is_a20x = is_a20x(fd_context(pctx)->screen);
52    return so;
53 }
54 
55 static void
delete_shader(struct fd2_shader_stateobj * so)56 delete_shader(struct fd2_shader_stateobj *so)
57 {
58    if (!so)
59       return;
60    ralloc_free(so->nir);
61    for (int i = 0; i < ARRAY_SIZE(so->variant); i++)
62       free(so->variant[i].info.dwords);
63    free(so);
64 }
65 
66 static void
emit(struct fd_ringbuffer * ring,gl_shader_stage type,struct ir2_shader_info * info,struct util_dynarray * patches)67 emit(struct fd_ringbuffer *ring, gl_shader_stage type,
68      struct ir2_shader_info *info, struct util_dynarray *patches)
69 {
70    unsigned i;
71 
72    assert(info->sizedwords);
73 
74    OUT_PKT3(ring, CP_IM_LOAD_IMMEDIATE, 2 + info->sizedwords);
75    OUT_RING(ring, type == MESA_SHADER_FRAGMENT);
76    OUT_RING(ring, info->sizedwords);
77 
78    if (patches)
79       util_dynarray_append(patches, uint32_t *,
80                            &ring->cur[info->mem_export_ptr]);
81 
82    for (i = 0; i < info->sizedwords; i++)
83       OUT_RING(ring, info->dwords[i]);
84 }
85 
86 static int
ir2_glsl_type_size(const struct glsl_type * type,bool bindless)87 ir2_glsl_type_size(const struct glsl_type *type, bool bindless)
88 {
89    return glsl_count_attribute_slots(type, false);
90 }
91 
92 static void *
fd2_fp_state_create(struct pipe_context * pctx,const struct pipe_shader_state * cso)93 fd2_fp_state_create(struct pipe_context *pctx,
94                     const struct pipe_shader_state *cso)
95 {
96    struct fd2_shader_stateobj *so = create_shader(pctx, MESA_SHADER_FRAGMENT);
97    if (!so)
98       return NULL;
99 
100    so->nir = (cso->type == PIPE_SHADER_IR_NIR)
101                 ? cso->ir.nir
102                 : tgsi_to_nir(cso->tokens, pctx->screen, false);
103 
104    NIR_PASS_V(so->nir, nir_lower_io, nir_var_shader_in | nir_var_shader_out,
105               ir2_glsl_type_size, (nir_lower_io_options)0);
106 
107    if (ir2_optimize_nir(so->nir, true))
108       goto fail;
109 
110    so->first_immediate = so->nir->num_uniforms;
111 
112    ir2_compile(so, 0, NULL);
113 
114    ralloc_free(so->nir);
115    so->nir = NULL;
116    return so;
117 
118 fail:
119    delete_shader(so);
120    return NULL;
121 }
122 
123 static void
fd2_fp_state_delete(struct pipe_context * pctx,void * hwcso)124 fd2_fp_state_delete(struct pipe_context *pctx, void *hwcso)
125 {
126    struct fd2_shader_stateobj *so = hwcso;
127    delete_shader(so);
128 }
129 
130 static void *
fd2_vp_state_create(struct pipe_context * pctx,const struct pipe_shader_state * cso)131 fd2_vp_state_create(struct pipe_context *pctx,
132                     const struct pipe_shader_state *cso)
133 {
134    struct fd2_shader_stateobj *so = create_shader(pctx, MESA_SHADER_VERTEX);
135    if (!so)
136       return NULL;
137 
138    so->nir = (cso->type == PIPE_SHADER_IR_NIR)
139                 ? cso->ir.nir
140                 : tgsi_to_nir(cso->tokens, pctx->screen, false);
141 
142    NIR_PASS_V(so->nir, nir_lower_io, nir_var_shader_in | nir_var_shader_out,
143               ir2_glsl_type_size, (nir_lower_io_options)0);
144 
145    if (ir2_optimize_nir(so->nir, true))
146       goto fail;
147 
148    so->first_immediate = so->nir->num_uniforms;
149 
150    /* compile binning variant now */
151    ir2_compile(so, 0, NULL);
152 
153    return so;
154 
155 fail:
156    delete_shader(so);
157    return NULL;
158 }
159 
160 static void
fd2_vp_state_delete(struct pipe_context * pctx,void * hwcso)161 fd2_vp_state_delete(struct pipe_context *pctx, void *hwcso)
162 {
163    struct fd2_shader_stateobj *so = hwcso;
164    delete_shader(so);
165 }
166 
167 static void
patch_vtx_fetch(struct fd_context * ctx,struct pipe_vertex_element * elem,instr_fetch_vtx_t * instr,uint16_t dst_swiz)168 patch_vtx_fetch(struct fd_context *ctx, struct pipe_vertex_element *elem,
169                 instr_fetch_vtx_t *instr, uint16_t dst_swiz) assert_dt
170 {
171    struct surface_format fmt = fd2_pipe2surface(elem->src_format);
172 
173    instr->dst_swiz = fd2_vtx_swiz(elem->src_format, dst_swiz);
174    instr->format_comp_all = fmt.sign == SQ_TEX_SIGN_SIGNED;
175    instr->num_format_all = fmt.num_format;
176    instr->format = fmt.format;
177    instr->exp_adjust_all = fmt.exp_adjust;
178    instr->stride = elem->src_stride;
179    instr->offset = elem->src_offset;
180 }
181 
182 static void
patch_fetches(struct fd_context * ctx,struct ir2_shader_info * info,struct fd_vertex_stateobj * vtx,struct fd_texture_stateobj * tex)183 patch_fetches(struct fd_context *ctx, struct ir2_shader_info *info,
184               struct fd_vertex_stateobj *vtx,
185               struct fd_texture_stateobj *tex) assert_dt
186 {
187    for (int i = 0; i < info->num_fetch_instrs; i++) {
188       struct ir2_fetch_info *fi = &info->fetch_info[i];
189 
190       instr_fetch_t *instr = (instr_fetch_t *)&info->dwords[fi->offset];
191       if (instr->opc == VTX_FETCH) {
192          unsigned idx =
193             (instr->vtx.const_index - 20) * 3 + instr->vtx.const_index_sel;
194          patch_vtx_fetch(ctx, &vtx->pipe[idx], &instr->vtx, fi->vtx.dst_swiz);
195          continue;
196       }
197 
198       assert(instr->opc == TEX_FETCH);
199       instr->tex.const_idx = fd2_get_const_idx(ctx, tex, fi->tex.samp_id);
200       instr->tex.src_swiz = fi->tex.src_swiz;
201    }
202 }
203 
204 void
fd2_program_emit(struct fd_context * ctx,struct fd_ringbuffer * ring,struct fd_program_stateobj * prog)205 fd2_program_emit(struct fd_context *ctx, struct fd_ringbuffer *ring,
206                  struct fd_program_stateobj *prog)
207 {
208    struct fd2_shader_stateobj *fp = NULL, *vp;
209    struct ir2_shader_info *fpi, *vpi;
210    struct ir2_frag_linkage *f;
211    uint8_t vs_gprs, fs_gprs = 0, vs_export = 0;
212    enum a2xx_sq_ps_vtx_mode mode = POSITION_1_VECTOR;
213    bool binning = (ctx->batch && ring == ctx->batch->binning);
214    unsigned variant = 0;
215 
216    vp = prog->vs;
217 
218    /* find variant matching the linked fragment shader */
219    if (!binning) {
220       fp = prog->fs;
221       for (variant = 1; variant < ARRAY_SIZE(vp->variant); variant++) {
222          /* if checked all variants, compile a new variant */
223          if (!vp->variant[variant].info.sizedwords) {
224             ir2_compile(vp, variant, fp);
225             break;
226          }
227 
228          /* check if fragment shader linkage matches */
229          if (!memcmp(&vp->variant[variant].f, &fp->variant[0].f,
230                      sizeof(struct ir2_frag_linkage)))
231             break;
232       }
233       assert(variant < ARRAY_SIZE(vp->variant));
234    }
235 
236    vpi = &vp->variant[variant].info;
237    fpi = &fp->variant[0].info;
238    f = &fp->variant[0].f;
239 
240    /* clear/gmem2mem/mem2gmem need to be changed to remove this condition */
241    if (prog != &ctx->solid_prog && prog != &ctx->blit_prog[0]) {
242       patch_fetches(ctx, vpi, ctx->vtx.vtx, &ctx->tex[PIPE_SHADER_VERTEX]);
243       if (fp)
244          patch_fetches(ctx, fpi, NULL, &ctx->tex[PIPE_SHADER_FRAGMENT]);
245    }
246 
247    emit(ring, MESA_SHADER_VERTEX, vpi,
248         binning ? &ctx->batch->shader_patches : NULL);
249 
250    if (fp) {
251       emit(ring, MESA_SHADER_FRAGMENT, fpi, NULL);
252       fs_gprs = (fpi->max_reg < 0) ? 0x80 : fpi->max_reg;
253       vs_export = MAX2(1, f->inputs_count) - 1;
254    }
255 
256    vs_gprs = (vpi->max_reg < 0) ? 0x80 : vpi->max_reg;
257 
258    if (vp->writes_psize && !binning)
259       mode = POSITION_2_VECTORS_SPRITE;
260 
261    /* set register to use for param (fragcoord/pointcoord/frontfacing) */
262    OUT_PKT3(ring, CP_SET_CONSTANT, 2);
263    OUT_RING(ring, CP_REG(REG_A2XX_SQ_CONTEXT_MISC));
264    OUT_RING(ring,
265             A2XX_SQ_CONTEXT_MISC_SC_SAMPLE_CNTL(CENTERS_ONLY) |
266                COND(fp, A2XX_SQ_CONTEXT_MISC_PARAM_GEN_POS(f->inputs_count)) |
267                /* we need SCREEN_XY for both fragcoord and frontfacing */
268                A2XX_SQ_CONTEXT_MISC_SC_OUTPUT_SCREEN_XY);
269 
270    OUT_PKT3(ring, CP_SET_CONSTANT, 2);
271    OUT_RING(ring, CP_REG(REG_A2XX_SQ_PROGRAM_CNTL));
272    OUT_RING(ring,
273             A2XX_SQ_PROGRAM_CNTL_PS_EXPORT_MODE(2) |
274                A2XX_SQ_PROGRAM_CNTL_VS_EXPORT_MODE(mode) |
275                A2XX_SQ_PROGRAM_CNTL_VS_RESOURCE |
276                A2XX_SQ_PROGRAM_CNTL_PS_RESOURCE |
277                A2XX_SQ_PROGRAM_CNTL_VS_EXPORT_COUNT(vs_export) |
278                A2XX_SQ_PROGRAM_CNTL_PS_REGS(fs_gprs) |
279                A2XX_SQ_PROGRAM_CNTL_VS_REGS(vs_gprs) |
280                COND(fp && fp->need_param, A2XX_SQ_PROGRAM_CNTL_PARAM_GEN) |
281                COND(!fp, A2XX_SQ_PROGRAM_CNTL_GEN_INDEX_VTX));
282 }
283 
284 void
fd2_prog_init(struct pipe_context * pctx)285 fd2_prog_init(struct pipe_context *pctx)
286 {
287    struct fd_context *ctx = fd_context(pctx);
288    struct fd_program_stateobj *prog;
289    struct fd2_shader_stateobj *so;
290    struct ir2_shader_info *info;
291    instr_fetch_vtx_t *instr;
292 
293    pctx->create_fs_state = fd2_fp_state_create;
294    pctx->delete_fs_state = fd2_fp_state_delete;
295 
296    pctx->create_vs_state = fd2_vp_state_create;
297    pctx->delete_vs_state = fd2_vp_state_delete;
298 
299    fd_prog_init(pctx);
300 
301    /* XXX maybe its possible to reuse patch_vtx_fetch somehow? */
302 
303    prog = &ctx->solid_prog;
304    so = prog->vs;
305    ir2_compile(prog->vs, 1, prog->fs);
306 
307 #define IR2_FETCH_SWIZ_XY01 0xb08
308 #define IR2_FETCH_SWIZ_XYZ1 0xa88
309 
310    info = &so->variant[1].info;
311 
312    instr = (instr_fetch_vtx_t *)&info->dwords[info->fetch_info[0].offset];
313    instr->const_index = 26;
314    instr->const_index_sel = 0;
315    instr->format = FMT_32_32_32_FLOAT;
316    instr->format_comp_all = false;
317    instr->stride = 12;
318    instr->num_format_all = true;
319    instr->dst_swiz = IR2_FETCH_SWIZ_XYZ1;
320 
321    prog = &ctx->blit_prog[0];
322    so = prog->vs;
323    ir2_compile(prog->vs, 1, prog->fs);
324 
325    info = &so->variant[1].info;
326 
327    instr = (instr_fetch_vtx_t *)&info->dwords[info->fetch_info[0].offset];
328    instr->const_index = 26;
329    instr->const_index_sel = 1;
330    instr->format = FMT_32_32_FLOAT;
331    instr->format_comp_all = false;
332    instr->stride = 8;
333    instr->num_format_all = false;
334    instr->dst_swiz = IR2_FETCH_SWIZ_XY01;
335 
336    instr = (instr_fetch_vtx_t *)&info->dwords[info->fetch_info[1].offset];
337    instr->const_index = 26;
338    instr->const_index_sel = 0;
339    instr->format = FMT_32_32_32_FLOAT;
340    instr->format_comp_all = false;
341    instr->stride = 12;
342    instr->num_format_all = false;
343    instr->dst_swiz = IR2_FETCH_SWIZ_XYZ1;
344 }
345