• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright © 2015 Red Hat
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 
24 #include "st_nir.h"
25 
26 #include "pipe/p_defines.h"
27 #include "pipe/p_screen.h"
28 #include "pipe/p_context.h"
29 
30 #include "program/program.h"
31 #include "program/prog_statevars.h"
32 #include "program/prog_parameter.h"
33 #include "program/ir_to_mesa.h"
34 #include "main/mtypes.h"
35 #include "main/errors.h"
36 #include "main/shaderapi.h"
37 #include "main/uniforms.h"
38 #include "util/string_to_uint_map.h"
39 
40 #include "st_context.h"
41 #include "st_program.h"
42 #include "st_glsl_types.h"
43 
44 #include "compiler/nir/nir.h"
45 #include "compiler/glsl_types.h"
46 #include "compiler/glsl/glsl_to_nir.h"
47 #include "compiler/glsl/ir.h"
48 
49 
50 /* Depending on PIPE_CAP_TGSI_TEXCOORD (st->needs_texcoord_semantic) we
51  * may need to fix up varying slots so the glsl->nir path is aligned
52  * with the anything->tgsi->nir path.
53  */
54 static void
st_nir_fixup_varying_slots(struct st_context * st,struct exec_list * var_list)55 st_nir_fixup_varying_slots(struct st_context *st, struct exec_list *var_list)
56 {
57    if (st->needs_texcoord_semantic)
58       return;
59 
60    nir_foreach_variable(var, var_list) {
61       if (var->data.location >= VARYING_SLOT_VAR0) {
62          var->data.location += 9;
63       } else if ((var->data.location >= VARYING_SLOT_TEX0) &&
64                (var->data.location <= VARYING_SLOT_TEX7)) {
65          var->data.location += VARYING_SLOT_VAR0 - VARYING_SLOT_TEX0;
66       }
67    }
68 }
69 
70 /* input location assignment for VS inputs must be handled specially, so
71  * that it is aligned w/ st's vbo state.
72  * (This isn't the case with, for ex, FS inputs, which only need to agree
73  * on varying-slot w/ the VS outputs)
74  */
75 static void
st_nir_assign_vs_in_locations(struct gl_program * prog,nir_shader * nir)76 st_nir_assign_vs_in_locations(struct gl_program *prog, nir_shader *nir)
77 {
78    unsigned attr, num_inputs = 0;
79    unsigned input_to_index[VERT_ATTRIB_MAX] = {0};
80 
81    /* TODO de-duplicate w/ similar code in st_translate_vertex_program()? */
82    for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
83       if ((prog->info.inputs_read & BITFIELD64_BIT(attr)) != 0) {
84          input_to_index[attr] = num_inputs;
85          num_inputs++;
86          if ((prog->info.double_inputs_read & BITFIELD64_BIT(attr)) != 0) {
87             /* add placeholder for second part of a double attribute */
88             num_inputs++;
89          }
90       } else {
91          input_to_index[attr] = ~0;
92       }
93    }
94 
95    nir->num_inputs = 0;
96    nir_foreach_variable_safe(var, &nir->inputs) {
97       attr = var->data.location;
98       assert(attr < ARRAY_SIZE(input_to_index));
99 
100       if (input_to_index[attr] != ~0u) {
101          var->data.driver_location = input_to_index[attr];
102          nir->num_inputs++;
103       } else {
104          /* Move unused input variables to the globals list (with no
105           * initialization), to avoid confusing drivers looking through the
106           * inputs array and expecting to find inputs with a driver_location
107           * set.
108           */
109          exec_node_remove(&var->node);
110          var->data.mode = nir_var_global;
111          exec_list_push_tail(&nir->globals, &var->node);
112       }
113    }
114 }
115 
116 static int
st_nir_lookup_parameter_index(const struct gl_program_parameter_list * params,const char * name)117 st_nir_lookup_parameter_index(const struct gl_program_parameter_list *params,
118                               const char *name)
119 {
120    int loc = _mesa_lookup_parameter_index(params, name);
121 
122    /* is there a better way to do this?  If we have something like:
123     *
124     *    struct S {
125     *           float f;
126     *           vec4 v;
127     *    };
128     *    uniform S color;
129     *
130     * Then what we get in prog->Parameters looks like:
131     *
132     *    0: Name=color.f, Type=6, DataType=1406, Size=1
133     *    1: Name=color.v, Type=6, DataType=8b52, Size=4
134     *
135     * So the name doesn't match up and _mesa_lookup_parameter_index()
136     * fails.  In this case just find the first matching "color.*"..
137     *
138     * Note for arrays you could end up w/ color[n].f, for example.
139     *
140     * glsl_to_tgsi works slightly differently in this regard.  It is
141     * emitting something more low level, so it just translates the
142     * params list 1:1 to CONST[] regs.  Going from GLSL IR to TGSI,
143     * it just calculates the additional offset of struct field members
144     * in glsl_to_tgsi_visitor::visit(ir_dereference_record *ir) or
145     * glsl_to_tgsi_visitor::visit(ir_dereference_array *ir).  It never
146     * needs to work backwards to get base var loc from the param-list
147     * which already has them separated out.
148     */
149    if (loc < 0) {
150       int namelen = strlen(name);
151       for (unsigned i = 0; i < params->NumParameters; i++) {
152          struct gl_program_parameter *p = &params->Parameters[i];
153          if ((strncmp(p->Name, name, namelen) == 0) &&
154              ((p->Name[namelen] == '.') || (p->Name[namelen] == '['))) {
155             loc = i;
156             break;
157          }
158       }
159    }
160 
161    return loc;
162 }
163 
164 static void
st_nir_assign_uniform_locations(struct gl_program * prog,struct gl_shader_program * shader_program,struct exec_list * uniform_list,unsigned * size)165 st_nir_assign_uniform_locations(struct gl_program *prog,
166                                 struct gl_shader_program *shader_program,
167                                 struct exec_list *uniform_list, unsigned *size)
168 {
169    int max = 0;
170    int shaderidx = 0;
171 
172    nir_foreach_variable(uniform, uniform_list) {
173       int loc;
174 
175       /*
176        * UBO's have their own address spaces, so don't count them towards the
177        * number of global uniforms
178        */
179       if ((uniform->data.mode == nir_var_uniform || uniform->data.mode == nir_var_shader_storage) &&
180           uniform->interface_type != NULL)
181          continue;
182 
183       if (uniform->type->is_sampler()) {
184          unsigned val;
185          bool found = shader_program->UniformHash->get(val, uniform->name);
186          loc = shaderidx++;
187          assert(found);
188          /* this ensure that nir_lower_samplers looks at the correct
189           * shader_program->UniformStorage[location]:
190           */
191          uniform->data.location = val;
192       } else if (strncmp(uniform->name, "gl_", 3) == 0) {
193          const gl_state_index *const stateTokens = (gl_state_index *)uniform->state_slots[0].tokens;
194          /* This state reference has already been setup by ir_to_mesa, but we'll
195           * get the same index back here.
196           */
197          loc = _mesa_add_state_reference(prog->Parameters, stateTokens);
198       } else {
199          loc = st_nir_lookup_parameter_index(prog->Parameters, uniform->name);
200       }
201 
202       uniform->data.driver_location = loc;
203 
204       max = MAX2(max, loc + st_glsl_type_size(uniform->type));
205    }
206    *size = max;
207 }
208 
209 extern "C" {
210 
211 /* First half of converting glsl_to_nir.. this leaves things in a pre-
212  * nir_lower_io state, so that shader variants can more easily insert/
213  * replace variables, etc.
214  */
215 nir_shader *
st_glsl_to_nir(struct st_context * st,struct gl_program * prog,struct gl_shader_program * shader_program,gl_shader_stage stage)216 st_glsl_to_nir(struct st_context *st, struct gl_program *prog,
217                struct gl_shader_program *shader_program,
218                gl_shader_stage stage)
219 {
220    struct pipe_screen *pscreen = st->pipe->screen;
221    enum pipe_shader_type ptarget = st_shader_stage_to_ptarget(stage);
222    const nir_shader_compiler_options *options;
223    nir_shader *nir;
224 
225    assert(pscreen->get_compiler_options);   /* drivers using NIR must implement this */
226 
227    options = (const nir_shader_compiler_options *)
228       pscreen->get_compiler_options(pscreen, PIPE_SHADER_IR_NIR, ptarget);
229    assert(options);
230 
231    if (prog->nir)
232       return prog->nir;
233 
234    nir = glsl_to_nir(shader_program, stage, options);
235    prog->nir = nir;
236 
237    NIR_PASS_V(nir, nir_lower_io_to_temporaries,
238          nir_shader_get_entrypoint(nir),
239          true, true);
240    NIR_PASS_V(nir, nir_lower_global_vars_to_local);
241    NIR_PASS_V(nir, nir_split_var_copies);
242    NIR_PASS_V(nir, nir_lower_var_copies);
243    NIR_PASS_V(nir, st_nir_lower_builtin);
244 
245    /* fragment shaders may need : */
246    if (stage == MESA_SHADER_FRAGMENT) {
247       static const gl_state_index wposTransformState[STATE_LENGTH] = {
248          STATE_INTERNAL, STATE_FB_WPOS_Y_TRANSFORM
249       };
250       nir_lower_wpos_ytransform_options wpos_options = {0};
251       struct pipe_screen *pscreen = st->pipe->screen;
252 
253       memcpy(wpos_options.state_tokens, wposTransformState,
254              sizeof(wpos_options.state_tokens));
255       wpos_options.fs_coord_origin_upper_left =
256          pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT);
257       wpos_options.fs_coord_origin_lower_left =
258          pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_ORIGIN_LOWER_LEFT);
259       wpos_options.fs_coord_pixel_center_integer =
260          pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_INTEGER);
261       wpos_options.fs_coord_pixel_center_half_integer =
262          pscreen->get_param(pscreen, PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER);
263 
264       if (nir_lower_wpos_ytransform(nir, &wpos_options)) {
265          nir_validate_shader(nir);
266          _mesa_add_state_reference(prog->Parameters, wposTransformState);
267       }
268    }
269 
270    if (st->ctx->_Shader->Flags & GLSL_DUMP) {
271       _mesa_log("\n");
272       _mesa_log("NIR IR for linked %s program %d:\n",
273              _mesa_shader_stage_to_string(stage),
274              shader_program->Name);
275       nir_print_shader(nir, _mesa_get_log_file());
276       _mesa_log("\n\n");
277    }
278 
279    return nir;
280 }
281 
282 /* TODO any better helper somewhere to sort a list? */
283 
284 static void
insert_sorted(struct exec_list * var_list,nir_variable * new_var)285 insert_sorted(struct exec_list *var_list, nir_variable *new_var)
286 {
287    nir_foreach_variable(var, var_list) {
288       if (var->data.location > new_var->data.location) {
289          exec_node_insert_node_before(&var->node, &new_var->node);
290          return;
291       }
292    }
293    exec_list_push_tail(var_list, &new_var->node);
294 }
295 
296 static void
sort_varyings(struct exec_list * var_list)297 sort_varyings(struct exec_list *var_list)
298 {
299    struct exec_list new_list;
300    exec_list_make_empty(&new_list);
301    nir_foreach_variable_safe(var, var_list) {
302       exec_node_remove(&var->node);
303       insert_sorted(&new_list, var);
304    }
305    exec_list_move_nodes_to(&new_list, var_list);
306 }
307 
308 /* Second half of preparing nir from glsl, which happens after shader
309  * variant lowering.
310  */
311 void
st_finalize_nir(struct st_context * st,struct gl_program * prog,nir_shader * nir)312 st_finalize_nir(struct st_context *st, struct gl_program *prog, nir_shader *nir)
313 {
314    NIR_PASS_V(nir, nir_split_var_copies);
315    NIR_PASS_V(nir, nir_lower_var_copies);
316    NIR_PASS_V(nir, nir_lower_io_types);
317 
318    if (nir->stage == MESA_SHADER_VERTEX) {
319       /* Needs special handling so drvloc matches the vbo state: */
320       st_nir_assign_vs_in_locations(prog, nir);
321       /* Re-lower global vars, to deal with any dead VS inputs. */
322       NIR_PASS_V(nir, nir_lower_global_vars_to_local);
323 
324       sort_varyings(&nir->outputs);
325       nir_assign_var_locations(&nir->outputs,
326                                &nir->num_outputs,
327                                st_glsl_type_size);
328       st_nir_fixup_varying_slots(st, &nir->outputs);
329    } else if (nir->stage == MESA_SHADER_FRAGMENT) {
330       sort_varyings(&nir->inputs);
331       nir_assign_var_locations(&nir->inputs,
332                                &nir->num_inputs,
333                                st_glsl_type_size);
334       st_nir_fixup_varying_slots(st, &nir->inputs);
335       nir_assign_var_locations(&nir->outputs,
336                                &nir->num_outputs,
337                                st_glsl_type_size);
338    } else {
339       unreachable("invalid shader type for tgsi bypass\n");
340    }
341 
342    struct gl_shader_program *shader_program;
343    switch (nir->stage) {
344    case MESA_SHADER_VERTEX:
345       shader_program = ((struct st_vertex_program *)prog)->shader_program;
346       break;
347    case MESA_SHADER_FRAGMENT:
348       shader_program = ((struct st_fragment_program *)prog)->shader_program;
349       break;
350    default:
351       assert(!"should not be reached");
352       return;
353    }
354 
355    st_nir_assign_uniform_locations(prog, shader_program,
356                                    &nir->uniforms, &nir->num_uniforms);
357 
358    NIR_PASS_V(nir, nir_lower_system_values);
359    NIR_PASS_V(nir, nir_lower_io, nir_var_all, st_glsl_type_size,
360               (nir_lower_io_options)0);
361    NIR_PASS_V(nir, nir_lower_samplers, shader_program);
362 }
363 
364 struct gl_program *
st_nir_get_mesa_program(struct gl_context * ctx,struct gl_shader_program * shader_program,struct gl_linked_shader * shader)365 st_nir_get_mesa_program(struct gl_context *ctx,
366                         struct gl_shader_program *shader_program,
367                         struct gl_linked_shader *shader)
368 {
369    struct gl_program *prog;
370 
371    validate_ir_tree(shader->ir);
372 
373    prog = shader->Program;
374 
375    prog->Parameters = _mesa_new_parameter_list();
376 
377    do_set_program_inouts(shader->ir, prog, shader->Stage);
378 
379    _mesa_copy_linked_program_data(shader_program, shader);
380    _mesa_generate_parameters_list_for_uniforms(shader_program, shader,
381                                                prog->Parameters);
382 
383    /* Make a pass over the IR to add state references for any built-in
384     * uniforms that are used.  This has to be done now (during linking).
385     * Code generation doesn't happen until the first time this shader is
386     * used for rendering.  Waiting until then to generate the parameters is
387     * too late.  At that point, the values for the built-in uniforms won't
388     * get sent to the shader.
389     */
390    foreach_in_list(ir_instruction, node, shader->ir) {
391       ir_variable *var = node->as_variable();
392 
393       if ((var == NULL) || (var->data.mode != ir_var_uniform) ||
394           (strncmp(var->name, "gl_", 3) != 0))
395          continue;
396 
397       const ir_state_slot *const slots = var->get_state_slots();
398       assert(slots != NULL);
399 
400       for (unsigned int i = 0; i < var->get_num_state_slots(); i++) {
401          _mesa_add_state_reference(prog->Parameters,
402                                    (gl_state_index *) slots[i].tokens);
403       }
404    }
405 
406    if (ctx->_Shader->Flags & GLSL_DUMP) {
407       _mesa_log("\n");
408       _mesa_log("GLSL IR for linked %s program %d:\n",
409              _mesa_shader_stage_to_string(shader->Stage),
410              shader_program->Name);
411       _mesa_print_ir(_mesa_get_log_file(), shader->ir, NULL);
412       _mesa_log("\n\n");
413    }
414 
415    prog->ShadowSamplers = shader->shadow_samplers;
416    prog->ExternalSamplersUsed = gl_external_samplers(prog);
417    _mesa_update_shader_textures_used(shader_program, prog);
418 
419    /* Avoid reallocation of the program parameter list, because the uniform
420     * storage is only associated with the original parameter list.
421     * This should be enough for Bitmap and DrawPixels constants.
422     */
423    _mesa_reserve_parameter_storage(prog->Parameters, 8);
424 
425    /* This has to be done last.  Any operation the can cause
426     * prog->ParameterValues to get reallocated (e.g., anything that adds a
427     * program constant) has to happen before creating this linkage.
428     */
429    _mesa_associate_uniform_storage(ctx, shader_program, prog->Parameters);
430 
431    struct st_vertex_program *stvp;
432    struct st_fragment_program *stfp;
433 
434    switch (shader->Stage) {
435    case MESA_SHADER_VERTEX:
436       stvp = (struct st_vertex_program *)prog;
437       stvp->shader_program = shader_program;
438       break;
439    case MESA_SHADER_FRAGMENT:
440       stfp = (struct st_fragment_program *)prog;
441       stfp->shader_program = shader_program;
442       break;
443    default:
444       assert(!"should not be reached");
445       return NULL;
446    }
447 
448    return prog;
449 }
450 
451 } /* extern "C" */
452