1 /*
2 * Copyright © 2018 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 DEALINGS
21 * IN THE SOFTWARE.
22 */
23
24 #include "nir.h"
25 #include "nir_gl_types.h"
26 #include "nir_xfb_info.h"
27 #include "gl_nir_linker.h"
28 #include "linker_util.h"
29 #include "util/u_math.h"
30 #include "main/shader_types.h"
31 #include "main/consts_exts.h"
32
33 /**
34 * This file does the linking of GLSL transform feedback using NIR.
35 */
36
37 void
gl_nir_link_assign_xfb_resources(const struct gl_constants * consts,struct gl_shader_program * prog)38 gl_nir_link_assign_xfb_resources(const struct gl_constants *consts,
39 struct gl_shader_program *prog)
40 {
41 /*
42 * From ARB_gl_spirv spec:
43 *
44 * "- If the *Xfb* Execution Mode is set, any output variable that is at
45 * least partially captured:
46 * * must be decorated with an *XfbBuffer*, declaring the capturing buffer
47 * * must have at least one captured output variable in the capturing
48 * buffer decorated with an *XfbStride* (and all such *XfbStride* values
49 * for the capturing buffer must be equal)
50 * - If the *Xfb* Execution Mode is set, any captured output:
51 * * must be a non-structure decorated with *Offset* or a member of a
52 * structure whose type member is decorated with *Offset*"
53 *
54 * Note the "must be", meaning that explicit buffer, offset and stride are
55 * mandatory. So as this is intended to work with SPIR-V shaders we don't
56 * need to calculate the offset or the stride.
57 */
58
59 struct gl_program *xfb_prog = prog->last_vert_prog;
60
61 if (xfb_prog == NULL)
62 return;
63
64 /* free existing varyings, if any */
65 for (unsigned i = 0; i < prog->TransformFeedback.NumVarying; i++)
66 free(prog->TransformFeedback.VaryingNames[i]);
67 free(prog->TransformFeedback.VaryingNames);
68
69 nir_xfb_info *xfb_info = NULL;
70 nir_xfb_varyings_info *varyings_info = NULL;
71
72 /* Find last stage before fragment shader */
73 for (int stage = MESA_SHADER_FRAGMENT - 1; stage >= 0; stage--) {
74 struct gl_linked_shader *sh = prog->_LinkedShaders[stage];
75
76 if (sh && stage != MESA_SHADER_TESS_CTRL) {
77 nir_gather_xfb_info_with_varyings(sh->Program->nir, NULL, &varyings_info);
78 xfb_info = sh->Program->nir->xfb_info;
79 break;
80 }
81 }
82
83 struct gl_transform_feedback_info *linked_xfb =
84 rzalloc(xfb_prog, struct gl_transform_feedback_info);
85 xfb_prog->sh.LinkedTransformFeedback = linked_xfb;
86
87 if (!xfb_info) {
88 prog->TransformFeedback.NumVarying = 0;
89 linked_xfb->NumOutputs = 0;
90 linked_xfb->NumVarying = 0;
91 linked_xfb->ActiveBuffers = 0;
92 return;
93 }
94
95 for (unsigned buf = 0; buf < MAX_FEEDBACK_BUFFERS; buf++)
96 prog->TransformFeedback.BufferStride[buf] = xfb_info->buffers[buf].stride;
97
98 prog->TransformFeedback.NumVarying = varyings_info->varying_count;
99 prog->TransformFeedback.VaryingNames =
100 malloc(sizeof(GLchar *) * varyings_info->varying_count);
101
102 linked_xfb->Outputs =
103 rzalloc_array(xfb_prog,
104 struct gl_transform_feedback_output,
105 xfb_info->output_count);
106 linked_xfb->NumOutputs = xfb_info->output_count;
107
108 linked_xfb->Varyings =
109 rzalloc_array(xfb_prog,
110 struct gl_transform_feedback_varying_info,
111 varyings_info->varying_count);
112 linked_xfb->NumVarying = varyings_info->varying_count;
113
114 int buffer_index = 0; /* Corresponds to GL_TRANSFORM_FEEDBACK_BUFFER_INDEX */
115 int xfb_buffer =
116 (varyings_info->varying_count > 0) ?
117 xfb_info->outputs[0].buffer : 0;
118
119 for (unsigned i = 0; i < varyings_info->varying_count; i++) {
120 nir_xfb_varying_info *xfb_varying = &varyings_info->varyings[i];
121
122 /* From ARB_gl_spirv spec:
123 *
124 * "19. How should the program interface query operations behave for
125 * program objects created from SPIR-V shaders?
126 *
127 * DISCUSSION: we previously said we didn't need reflection to work
128 * for SPIR-V shaders (at least for the first version), however we
129 * are left with specifying how it should "not work". The primary
130 * issue is that SPIR-V binaries are not required to have names
131 * associated with variables. They can be associated in debug
132 * information, but there is no requirement for that to be present,
133 * and it should not be relied upon."
134 *
135 * Options:"
136 *
137 * <skip>
138 *
139 * "RESOLVED. Pick (c), but also allow debug names to be returned
140 * if an implementation wants to."
141 *
142 * So names are considered optional debug info, so the linker needs to
143 * work without them, and returning them is optional. For simplicity at
144 * this point we are ignoring names
145 */
146 prog->TransformFeedback.VaryingNames[i] = NULL;
147
148 if (xfb_buffer != xfb_varying->buffer) {
149 buffer_index++;
150 xfb_buffer = xfb_varying->buffer;
151 }
152
153 struct gl_transform_feedback_varying_info *varying =
154 linked_xfb->Varyings + i;
155
156 /* ARB_gl_spirv: see above. */
157 varying->name.string = NULL;
158 resource_name_updated(&varying->name);
159 varying->Type = glsl_get_gl_type(xfb_varying->type);
160 varying->BufferIndex = buffer_index;
161 varying->Size = glsl_type_is_array(xfb_varying->type) ?
162 glsl_get_length(xfb_varying->type) : 1;
163 varying->Offset = xfb_varying->offset;
164 }
165
166 for (unsigned i = 0; i < xfb_info->output_count; i++) {
167 nir_xfb_output_info *xfb_output = &xfb_info->outputs[i];
168
169 struct gl_transform_feedback_output *output =
170 linked_xfb->Outputs + i;
171
172 output->OutputRegister = xfb_output->location;
173 output->OutputBuffer = xfb_output->buffer;
174 output->NumComponents = util_bitcount(xfb_output->component_mask);
175 output->StreamId = xfb_info->buffer_to_stream[xfb_output->buffer];
176 output->DstOffset = xfb_output->offset / 4;
177 output->ComponentOffset = xfb_output->component_offset;
178 }
179
180 /* Make sure MaxTransformFeedbackBuffers is <= 32 so the bitmask for
181 * tracking the number of buffers doesn't overflow.
182 */
183 unsigned buffers = 0;
184 assert(consts->MaxTransformFeedbackBuffers <= sizeof(buffers) * 8);
185
186 for (unsigned buf = 0; buf < MAX_FEEDBACK_BUFFERS; buf++) {
187 if (xfb_info->buffers[buf].stride > 0) {
188 linked_xfb->Buffers[buf].Stride = xfb_info->buffers[buf].stride / 4;
189 linked_xfb->Buffers[buf].NumVaryings = xfb_info->buffers[buf].varying_count;
190 buffers |= 1 << buf;
191 }
192 }
193
194 linked_xfb->ActiveBuffers = buffers;
195
196 ralloc_free(varyings_info);
197 }
198
199 struct nir_xfb_info *
gl_to_nir_xfb_info(struct gl_transform_feedback_info * info,void * mem_ctx)200 gl_to_nir_xfb_info(struct gl_transform_feedback_info *info, void *mem_ctx)
201 {
202 if (info == NULL || info->NumOutputs == 0)
203 return NULL;
204
205 nir_xfb_info *xfb =
206 rzalloc_size(mem_ctx, nir_xfb_info_size(info->NumOutputs));
207
208 xfb->output_count = info->NumOutputs;
209
210 for (unsigned i = 0; i < MAX_FEEDBACK_BUFFERS; i++) {
211 xfb->buffers[i].stride = info->Buffers[i].Stride * 4;
212 xfb->buffers[i].varying_count = info->Buffers[i].NumVaryings;
213 xfb->buffer_to_stream[i] = info->Buffers[i].Stream;
214 }
215
216 for (unsigned i = 0; i < info->NumOutputs; i++) {
217 xfb->outputs[i].buffer = info->Outputs[i].OutputBuffer;
218 xfb->outputs[i].offset = info->Outputs[i].DstOffset * 4;
219 xfb->outputs[i].location = info->Outputs[i].OutputRegister;
220 xfb->outputs[i].component_offset = info->Outputs[i].ComponentOffset;
221 xfb->outputs[i].component_mask =
222 BITFIELD_RANGE(info->Outputs[i].ComponentOffset,
223 info->Outputs[i].NumComponents);
224 xfb->buffers_written |= BITFIELD_BIT(info->Outputs[i].OutputBuffer);
225 xfb->streams_written |= BITFIELD_BIT(info->Outputs[i].StreamId);
226 }
227
228 return xfb;
229 }
230