xref: /aosp_15_r20/external/mesa3d/src/compiler/nir/nir_passthrough_gs.c (revision 6104692788411f58d303aa86923a9ff6ecaded22)
1 /*
2  * Copyright © 2022 Collabora Ltc.
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 "util/ralloc.h"
25 #include "util/u_memory.h"
26 #include "nir.h"
27 #include "nir_builder.h"
28 #include "nir_xfb_info.h"
29 
30 static enum mesa_prim
gs_in_prim_for_topology(enum mesa_prim prim)31 gs_in_prim_for_topology(enum mesa_prim prim)
32 {
33    switch (prim) {
34    case MESA_PRIM_QUADS:
35       return MESA_PRIM_LINES_ADJACENCY;
36    default:
37       return u_decomposed_prim(prim);
38    }
39 }
40 
41 static enum mesa_prim
gs_out_prim_for_topology(enum mesa_prim prim)42 gs_out_prim_for_topology(enum mesa_prim prim)
43 {
44    switch (prim) {
45    case MESA_PRIM_POINTS:
46       return MESA_PRIM_POINTS;
47    case MESA_PRIM_LINES:
48    case MESA_PRIM_LINE_LOOP:
49    case MESA_PRIM_LINES_ADJACENCY:
50    case MESA_PRIM_LINE_STRIP_ADJACENCY:
51    case MESA_PRIM_LINE_STRIP:
52       return MESA_PRIM_LINE_STRIP;
53    case MESA_PRIM_TRIANGLES:
54    case MESA_PRIM_TRIANGLE_STRIP:
55    case MESA_PRIM_TRIANGLE_FAN:
56    case MESA_PRIM_TRIANGLES_ADJACENCY:
57    case MESA_PRIM_TRIANGLE_STRIP_ADJACENCY:
58    case MESA_PRIM_POLYGON:
59       return MESA_PRIM_TRIANGLE_STRIP;
60    case MESA_PRIM_QUADS:
61    case MESA_PRIM_QUAD_STRIP:
62    case MESA_PRIM_PATCHES:
63    default:
64       return MESA_PRIM_QUADS;
65    }
66 }
67 
68 static unsigned int
vertices_for_prim(enum mesa_prim prim)69 vertices_for_prim(enum mesa_prim prim)
70 {
71    switch (prim) {
72    case MESA_PRIM_POINTS:
73       return 1;
74    case MESA_PRIM_LINES:
75    case MESA_PRIM_LINE_LOOP:
76    case MESA_PRIM_LINES_ADJACENCY:
77    case MESA_PRIM_LINE_STRIP_ADJACENCY:
78    case MESA_PRIM_LINE_STRIP:
79       return 2;
80    case MESA_PRIM_TRIANGLES:
81    case MESA_PRIM_TRIANGLE_STRIP:
82    case MESA_PRIM_TRIANGLE_FAN:
83    case MESA_PRIM_TRIANGLES_ADJACENCY:
84    case MESA_PRIM_TRIANGLE_STRIP_ADJACENCY:
85    case MESA_PRIM_POLYGON:
86       return 3;
87    case MESA_PRIM_QUADS:
88    case MESA_PRIM_QUAD_STRIP:
89       return 4;
90    case MESA_PRIM_PATCHES:
91    default:
92       unreachable("unsupported primitive for gs input");
93    }
94 }
95 
96 static void
copy_vars(nir_builder * b,nir_deref_instr * dst,nir_deref_instr * src)97 copy_vars(nir_builder *b, nir_deref_instr *dst, nir_deref_instr *src)
98 {
99    assert(glsl_get_bare_type(dst->type) == glsl_get_bare_type(src->type));
100    if (glsl_type_is_struct_or_ifc(dst->type)) {
101       for (unsigned i = 0; i < glsl_get_length(dst->type); ++i) {
102          copy_vars(b, nir_build_deref_struct(b, dst, i), nir_build_deref_struct(b, src, i));
103       }
104    } else if (glsl_type_is_array_or_matrix(dst->type)) {
105       unsigned count = glsl_type_is_array(dst->type) ? glsl_array_size(dst->type) : glsl_get_matrix_columns(dst->type);
106       for (unsigned i = 0; i < count; i++) {
107          copy_vars(b, nir_build_deref_array_imm(b, dst, i), nir_build_deref_array_imm(b, src, i));
108       }
109    } else {
110       nir_def *load = nir_load_deref(b, src);
111       nir_store_deref(b, dst, load, BITFIELD_MASK(load->num_components));
112    }
113 }
114 
115 /*
116  * A helper to create a passthrough GS shader for drivers that needs to lower
117  * some rendering tasks to the GS.
118  */
119 
120 nir_shader *
nir_create_passthrough_gs(const nir_shader_compiler_options * options,const nir_shader * prev_stage,enum mesa_prim primitive_type,enum mesa_prim output_primitive_type,bool emulate_edgeflags,bool force_line_strip_out)121 nir_create_passthrough_gs(const nir_shader_compiler_options *options,
122                           const nir_shader *prev_stage,
123                           enum mesa_prim primitive_type,
124                           enum mesa_prim output_primitive_type,
125                           bool emulate_edgeflags,
126                           bool force_line_strip_out)
127 {
128    unsigned int vertices_out = vertices_for_prim(primitive_type);
129    emulate_edgeflags = emulate_edgeflags && (prev_stage->info.outputs_written & VARYING_BIT_EDGE);
130    enum mesa_prim original_our_prim = gs_out_prim_for_topology(output_primitive_type);
131    nir_builder b = nir_builder_init_simple_shader(MESA_SHADER_GEOMETRY,
132                                                   options,
133                                                   "gs passthrough");
134 
135    bool output_lines = force_line_strip_out || original_our_prim == MESA_PRIM_LINE_STRIP;
136    bool needs_closing = (force_line_strip_out || (emulate_edgeflags && output_lines)) && vertices_out >= 3;
137 
138    nir_shader *nir = b.shader;
139    nir->info.gs.input_primitive = gs_in_prim_for_topology(primitive_type);
140    nir->info.gs.output_primitive = force_line_strip_out ? MESA_PRIM_LINE_STRIP : original_our_prim;
141    nir->info.gs.vertices_in = mesa_vertices_per_prim(primitive_type);
142    nir->info.gs.vertices_out = needs_closing ? vertices_out + 1 : vertices_out;
143    nir->info.gs.invocations = 1;
144    nir->info.gs.active_stream_mask = 1;
145 
146    nir->info.has_transform_feedback_varyings = prev_stage->info.has_transform_feedback_varyings;
147    memcpy(nir->info.xfb_stride, prev_stage->info.xfb_stride, sizeof(prev_stage->info.xfb_stride));
148    if (prev_stage->xfb_info) {
149       size_t size = nir_xfb_info_size(prev_stage->xfb_info->output_count);
150       nir->xfb_info = ralloc_memdup(nir, prev_stage->xfb_info, size);
151    }
152 
153    bool handle_flat = output_lines && nir->info.gs.output_primitive != gs_out_prim_for_topology(primitive_type);
154    nir_variable *in_vars[VARYING_SLOT_MAX * 4];
155    nir_variable *out_vars[VARYING_SLOT_MAX * 4];
156    unsigned num_inputs = 0, num_outputs = 0;
157 
158    /* Create input/output variables. */
159    nir_foreach_shader_out_variable(var, prev_stage) {
160       assert(!var->data.patch);
161 
162       /* input vars can't be created for those */
163       if (var->data.location == VARYING_SLOT_LAYER ||
164           var->data.location == VARYING_SLOT_VIEW_INDEX)
165          continue;
166 
167       char name[100];
168       if (var->name)
169          snprintf(name, sizeof(name), "in_%s", var->name);
170       else
171          snprintf(name, sizeof(name), "in_%d", var->data.driver_location);
172 
173       nir_variable *in = nir_variable_clone(var, nir);
174       ralloc_free(in->name);
175       in->name = ralloc_strdup(in, name);
176       in->type = glsl_array_type(var->type, 6, false);
177       in->data.mode = nir_var_shader_in;
178       nir_shader_add_variable(nir, in);
179 
180       in_vars[num_inputs++] = in;
181 
182       nir->num_inputs++;
183       if (in->data.location == VARYING_SLOT_EDGE)
184          continue;
185 
186       if (var->data.location != VARYING_SLOT_POS)
187          nir->num_outputs++;
188 
189       if (var->name)
190          snprintf(name, sizeof(name), "out_%s", var->name);
191       else
192          snprintf(name, sizeof(name), "out_%d", var->data.driver_location);
193 
194       nir_variable *out = nir_variable_clone(var, nir);
195       ralloc_free(out->name);
196       out->name = ralloc_strdup(out, name);
197       out->data.mode = nir_var_shader_out;
198       nir_shader_add_variable(nir, out);
199 
200       out_vars[num_outputs++] = out;
201    }
202 
203    unsigned int start_vert = 0;
204    unsigned int end_vert = vertices_out;
205    unsigned int vert_step = 1;
206    switch (primitive_type) {
207    case MESA_PRIM_LINES_ADJACENCY:
208    case MESA_PRIM_LINE_STRIP_ADJACENCY:
209       start_vert = 1;
210       end_vert += 1;
211       break;
212    case MESA_PRIM_TRIANGLES_ADJACENCY:
213    case MESA_PRIM_TRIANGLE_STRIP_ADJACENCY:
214       end_vert = 5;
215       vert_step = 2;
216       break;
217    default:
218       break;
219    }
220 
221    nir_variable *edge_var = nir_find_variable_with_location(nir, nir_var_shader_in, VARYING_SLOT_EDGE);
222    nir_def *flat_interp_mask_def = nir_load_flat_mask(&b);
223    nir_def *last_pv_vert_def = nir_load_provoking_last(&b);
224    last_pv_vert_def = nir_ine_imm(&b, last_pv_vert_def, 0);
225    nir_def *start_vert_index = nir_imm_int(&b, start_vert);
226    nir_def *end_vert_index = nir_imm_int(&b, end_vert - 1);
227    nir_def *pv_vert_index = nir_bcsel(&b, last_pv_vert_def, end_vert_index, start_vert_index);
228    for (unsigned i = start_vert; i < end_vert || needs_closing; i += vert_step) {
229       int idx = i < end_vert ? i : start_vert;
230       /* Copy inputs to outputs. */
231       for (unsigned j = 0, oj = 0; j < num_inputs; ++j) {
232          if (in_vars[j]->data.location == VARYING_SLOT_EDGE) {
233             continue;
234          }
235          /* no need to use copy_var to save a lower pass */
236          nir_def *index;
237          if (in_vars[j]->data.location == VARYING_SLOT_POS || !handle_flat)
238             index = nir_imm_int(&b, idx);
239          else {
240             uint64_t mask = BITFIELD64_BIT(in_vars[j]->data.location);
241             index = nir_bcsel(&b, nir_ieq_imm(&b, nir_iand_imm(&b, flat_interp_mask_def, mask), 0), nir_imm_int(&b, idx), pv_vert_index);
242          }
243          nir_deref_instr *value = nir_build_deref_array(&b, nir_build_deref_var(&b, in_vars[j]), index);
244          copy_vars(&b, nir_build_deref_var(&b, out_vars[oj]), value);
245          ++oj;
246       }
247 
248       if (emulate_edgeflags && !output_lines) {
249          nir_def *edge_value = nir_channel(&b, nir_load_array_var_imm(&b, edge_var, idx), 0);
250          nir_push_if(&b, nir_feq_imm(&b, edge_value, 1.0));
251       }
252 
253       nir_emit_vertex(&b, 0);
254 
255       if (emulate_edgeflags) {
256          if (nir->info.gs.output_primitive == MESA_PRIM_LINE_STRIP) {
257             nir_def *edge_value = nir_channel(&b, nir_load_array_var_imm(&b, edge_var, idx), 0);
258             nir_push_if(&b, nir_fneu_imm(&b, edge_value, 1.0));
259             nir_end_primitive(&b, 0);
260 
261          }
262          nir_pop_if(&b, NULL);
263       }
264 
265       if (i >= end_vert)
266          break;
267    }
268 
269    nir_end_primitive(&b, 0);
270    nir_shader_gather_info(nir, nir_shader_get_entrypoint(nir));
271    nir_validate_shader(nir, "in nir_create_passthrough_gs");
272 
273    return nir;
274 }
275