1 /*
2  * Copyright © 2013 Intel Corporation
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
20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
21  * DEALINGS IN THE SOFTWARE.
22  */
23 
24 /**
25  * \file brw_vec4_gs.c
26  *
27  * State atom for client-programmable geometry shaders, and support code.
28  */
29 
30 #include "brw_gs.h"
31 #include "brw_context.h"
32 #include "brw_vec4_gs_visitor.h"
33 #include "brw_state.h"
34 #include "brw_ff_gs.h"
35 #include "brw_nir.h"
36 #include "brw_program.h"
37 #include "compiler/glsl/ir_uniform.h"
38 
39 static void
40 brw_gs_debug_recompile(struct brw_context *brw, struct gl_program *prog,
41                        const struct brw_gs_prog_key *key)
42 {
43    perf_debug("Recompiling geometry shader for program %d\n", prog->Id);
44 
45    bool found = false;
46    const struct brw_gs_prog_key *old_key =
47       brw_find_previous_compile(&brw->cache, BRW_CACHE_GS_PROG,
48                                 key->program_string_id);
49 
50    if (!old_key) {
51       perf_debug("  Didn't find previous compile in the shader cache for "
52                  "debug\n");
53       return;
54    }
55 
56    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
57 
58    if (!found) {
59       perf_debug("  Something else\n");
60    }
61 }
62 
63 static void
64 assign_gs_binding_table_offsets(const struct gen_device_info *devinfo,
65                                 const struct gl_program *prog,
66                                 struct brw_gs_prog_data *prog_data)
67 {
68    /* In gen6 we reserve the first BRW_MAX_SOL_BINDINGS entries for transform
69     * feedback surfaces.
70     */
71    uint32_t reserved = devinfo->gen == 6 ? BRW_MAX_SOL_BINDINGS : 0;
72 
73    brw_assign_common_binding_table_offsets(devinfo, prog,
74                                            &prog_data->base.base, reserved);
75 }
76 
77 static bool
78 brw_codegen_gs_prog(struct brw_context *brw,
79                     struct brw_program *gp,
80                     struct brw_gs_prog_key *key)
81 {
82    struct brw_compiler *compiler = brw->screen->compiler;
83    const struct gen_device_info *devinfo = &brw->screen->devinfo;
84    struct brw_stage_state *stage_state = &brw->gs.base;
85    struct brw_gs_prog_data prog_data;
86    bool start_busy = false;
87    double start_time = 0;
88 
89    memset(&prog_data, 0, sizeof(prog_data));
90 
91    assign_gs_binding_table_offsets(devinfo, &gp->program, &prog_data);
92 
93    /* Allocate the references to the uniforms that will end up in the
94     * prog_data associated with the compiled program, and which will be freed
95     * by the state cache.
96     *
97     * Note: param_count needs to be num_uniform_components * 4, since we add
98     * padding around uniform values below vec4 size, so the worst case is that
99     * every uniform is a float which gets padded to the size of a vec4.
100     */
101    int param_count = gp->program.nir->num_uniforms / 4;
102 
103    prog_data.base.base.param =
104       rzalloc_array(NULL, const gl_constant_value *, param_count);
105    prog_data.base.base.pull_param =
106       rzalloc_array(NULL, const gl_constant_value *, param_count);
107    prog_data.base.base.image_param =
108       rzalloc_array(NULL, struct brw_image_param,
109                     gp->program.info.num_images);
110    prog_data.base.base.nr_params = param_count;
111    prog_data.base.base.nr_image_params = gp->program.info.num_images;
112 
113    brw_nir_setup_glsl_uniforms(gp->program.nir, &gp->program,
114                                &prog_data.base.base,
115                                compiler->scalar_stage[MESA_SHADER_GEOMETRY]);
116 
117    uint64_t outputs_written = gp->program.info.outputs_written;
118 
119    brw_compute_vue_map(devinfo,
120                        &prog_data.base.vue_map, outputs_written,
121                        gp->program.info.separate_shader);
122 
123    int st_index = -1;
124    if (INTEL_DEBUG & DEBUG_SHADER_TIME)
125       st_index = brw_get_shader_time_index(brw, &gp->program, ST_GS, true);
126 
127    if (unlikely(brw->perf_debug)) {
128       start_busy = brw->batch.last_bo && drm_intel_bo_busy(brw->batch.last_bo);
129       start_time = get_time();
130    }
131 
132    void *mem_ctx = ralloc_context(NULL);
133    unsigned program_size;
134    char *error_str;
135    const unsigned *program =
136       brw_compile_gs(brw->screen->compiler, brw, mem_ctx, key,
137                      &prog_data, gp->program.nir, &gp->program,
138                      st_index, &program_size, &error_str);
139    if (program == NULL) {
140       ralloc_strcat(&gp->program.sh.data->InfoLog, error_str);
141       _mesa_problem(NULL, "Failed to compile geometry shader: %s\n", error_str);
142 
143       ralloc_free(mem_ctx);
144       return false;
145    }
146 
147    if (unlikely(brw->perf_debug)) {
148       if (gp->compiled_once) {
149          brw_gs_debug_recompile(brw, &gp->program, key);
150       }
151       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
152          perf_debug("GS compile took %.03f ms and stalled the GPU\n",
153                     (get_time() - start_time) * 1000);
154       }
155       gp->compiled_once = true;
156    }
157 
158    /* Scratch space is used for register spilling */
159    brw_alloc_stage_scratch(brw, stage_state,
160                            prog_data.base.base.total_scratch,
161                            devinfo->max_gs_threads);
162 
163    brw_upload_cache(&brw->cache, BRW_CACHE_GS_PROG,
164                     key, sizeof(*key),
165                     program, program_size,
166                     &prog_data, sizeof(prog_data),
167                     &stage_state->prog_offset, &brw->gs.base.prog_data);
168    ralloc_free(mem_ctx);
169 
170    return true;
171 }
172 
173 static bool
174 brw_gs_state_dirty(const struct brw_context *brw)
175 {
176    return brw_state_dirty(brw,
177                           _NEW_TEXTURE,
178                           BRW_NEW_GEOMETRY_PROGRAM |
179                           BRW_NEW_TRANSFORM_FEEDBACK);
180 }
181 
182 void
183 brw_gs_populate_key(struct brw_context *brw,
184                     struct brw_gs_prog_key *key)
185 {
186    struct gl_context *ctx = &brw->ctx;
187    struct brw_program *gp = (struct brw_program *) brw->geometry_program;
188 
189    memset(key, 0, sizeof(*key));
190 
191    key->program_string_id = gp->id;
192 
193    /* _NEW_TEXTURE */
194    brw_populate_sampler_prog_key_data(ctx, &gp->program, &key->tex);
195 }
196 
197 void
198 brw_upload_gs_prog(struct brw_context *brw)
199 {
200    struct brw_stage_state *stage_state = &brw->gs.base;
201    struct brw_gs_prog_key key;
202    /* BRW_NEW_GEOMETRY_PROGRAM */
203    struct brw_program *gp = (struct brw_program *) brw->geometry_program;
204 
205    if (!brw_gs_state_dirty(brw))
206       return;
207 
208    if (gp == NULL) {
209       /* No geometry shader.  Vertex data just passes straight through. */
210       if (brw->gen == 6 &&
211           (brw->ctx.NewDriverState & BRW_NEW_TRANSFORM_FEEDBACK)) {
212          gen6_brw_upload_ff_gs_prog(brw);
213          return;
214       }
215 
216       /* Other state atoms had better not try to access prog_data, since
217        * there's no GS program.
218        */
219       brw->gs.base.prog_data = NULL;
220 
221       return;
222    }
223 
224    brw_gs_populate_key(brw, &key);
225 
226    if (!brw_search_cache(&brw->cache, BRW_CACHE_GS_PROG,
227                          &key, sizeof(key),
228                          &stage_state->prog_offset,
229                          &brw->gs.base.prog_data)) {
230       bool success = brw_codegen_gs_prog(brw, gp, &key);
231       assert(success);
232       (void)success;
233    }
234 }
235 
236 bool
237 brw_gs_precompile(struct gl_context *ctx, struct gl_program *prog)
238 {
239    struct brw_context *brw = brw_context(ctx);
240    struct brw_gs_prog_key key;
241    uint32_t old_prog_offset = brw->gs.base.prog_offset;
242    struct brw_stage_prog_data *old_prog_data = brw->gs.base.prog_data;
243    bool success;
244 
245    struct brw_program *bgp = brw_program(prog);
246 
247    memset(&key, 0, sizeof(key));
248 
249    brw_setup_tex_for_precompile(brw, &key.tex, prog);
250    key.program_string_id = bgp->id;
251 
252    success = brw_codegen_gs_prog(brw, bgp, &key);
253 
254    brw->gs.base.prog_offset = old_prog_offset;
255    brw->gs.base.prog_data = old_prog_data;
256 
257    return success;
258 }
259