1 /*
2 * Copyright 2012 Red Hat Inc.
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 shall be included in
12 * all copies or substantial portions of the Software.
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
17 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
21 *
22 * Authors: Ben Skeggs
23 *
24 */
25
26 #include "draw/draw_context.h"
27 #include "tgsi/tgsi_parse.h"
28
29 #include "nv_object.xml.h"
30 #include "nv30/nv30-40_3d.xml.h"
31 #include "nv30/nv30_context.h"
32 #include "nv30/nvfx_shader.h"
33
34 static void
nv30_fragprog_upload(struct nv30_context * nv30)35 nv30_fragprog_upload(struct nv30_context *nv30)
36 {
37 struct nouveau_context *nv = &nv30->base;
38 struct nv30_fragprog *fp = nv30->fragprog.program;
39 struct pipe_context *pipe = &nv30->base.pipe;
40
41 if (unlikely(!fp->buffer))
42 fp->buffer = pipe_buffer_create(pipe->screen, 0, 0, fp->insn_len * 4);
43
44 #ifndef PIPE_ARCH_BIG_ENDIAN
45 pipe_buffer_write(pipe, fp->buffer, 0, fp->insn_len * 4, fp->insn);
46 #else
47 {
48 struct pipe_transfer *transfer;
49 uint32_t *map;
50 int i;
51
52 map = pipe_buffer_map(pipe, fp->buffer,
53 PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD_WHOLE_RESOURCE,
54 &transfer);
55 for (i = 0; i < fp->insn_len; i++)
56 *map++ = (fp->insn[i] >> 16) | (fp->insn[i] << 16);
57 pipe_buffer_unmap(pipe, transfer);
58 }
59 #endif
60
61 if (nv04_resource(fp->buffer)->domain != NOUVEAU_BO_VRAM)
62 nouveau_buffer_migrate(nv, nv04_resource(fp->buffer), NOUVEAU_BO_VRAM);
63 }
64
65 void
nv30_fragprog_validate(struct nv30_context * nv30)66 nv30_fragprog_validate(struct nv30_context *nv30)
67 {
68 struct nouveau_pushbuf *push = nv30->base.pushbuf;
69 struct nouveau_object *eng3d = nv30->screen->eng3d;
70 struct nv30_fragprog *fp = nv30->fragprog.program;
71 bool upload = false;
72 int i;
73
74 if (!fp->translated) {
75 _nvfx_fragprog_translate(eng3d->oclass, fp);
76 if (!fp->translated)
77 return;
78
79 upload = true;
80 }
81
82 /* update constants, also needs to be done on every fp switch as we
83 * have no idea whether the constbuf changed in the meantime
84 */
85 if (nv30->fragprog.constbuf) {
86 struct pipe_resource *constbuf = nv30->fragprog.constbuf;
87 uint32_t *cbuf = (uint32_t *)nv04_resource(constbuf)->data;
88
89 for (i = 0; i < fp->nr_consts; i++) {
90 unsigned off = fp->consts[i].offset;
91 unsigned idx = fp->consts[i].index * 4;
92
93 if (!memcmp(&fp->insn[off], &cbuf[idx], 4 * 4))
94 continue;
95 memcpy(&fp->insn[off], &cbuf[idx], 4 * 4);
96 upload = true;
97 }
98 }
99
100 if (upload)
101 nv30_fragprog_upload(nv30);
102
103 /* FP_ACTIVE_PROGRAM needs to be done again even if only the consts
104 * were updated. TEX_CACHE_CTL magic is not enough to convince the
105 * GPU that it should re-read the fragprog from VRAM... sigh.
106 */
107 if (nv30->state.fragprog != fp || upload) {
108 struct nv04_resource *r = nv04_resource(fp->buffer);
109
110 if (!PUSH_SPACE(push, 8))
111 return;
112 PUSH_RESET(push, BUFCTX_FRAGPROG);
113
114 BEGIN_NV04(push, NV30_3D(FP_ACTIVE_PROGRAM), 1);
115 PUSH_RESRC(push, NV30_3D(FP_ACTIVE_PROGRAM), BUFCTX_FRAGPROG, r, 0,
116 NOUVEAU_BO_LOW | NOUVEAU_BO_RD | NOUVEAU_BO_OR,
117 NV30_3D_FP_ACTIVE_PROGRAM_DMA0,
118 NV30_3D_FP_ACTIVE_PROGRAM_DMA1);
119 BEGIN_NV04(push, NV30_3D(FP_CONTROL), 1);
120 PUSH_DATA (push, fp->fp_control);
121 if (eng3d->oclass < NV40_3D_CLASS) {
122 BEGIN_NV04(push, NV30_3D(FP_REG_CONTROL), 1);
123 PUSH_DATA (push, 0x00010004);
124 BEGIN_NV04(push, NV30_3D(TEX_UNITS_ENABLE), 1);
125 PUSH_DATA (push, fp->texcoords);
126 } else {
127 BEGIN_NV04(push, SUBC_3D(0x0b40), 1);
128 PUSH_DATA (push, 0x00000000);
129 }
130
131 nv30->state.fragprog = fp;
132 }
133 }
134
135 static void *
nv30_fp_state_create(struct pipe_context * pipe,const struct pipe_shader_state * cso)136 nv30_fp_state_create(struct pipe_context *pipe,
137 const struct pipe_shader_state *cso)
138 {
139 struct nv30_fragprog *fp = CALLOC_STRUCT(nv30_fragprog);
140 if (!fp)
141 return NULL;
142
143 fp->pipe.tokens = tgsi_dup_tokens(cso->tokens);
144 tgsi_scan_shader(fp->pipe.tokens, &fp->info);
145 return fp;
146 }
147
148 static void
nv30_fp_state_delete(struct pipe_context * pipe,void * hwcso)149 nv30_fp_state_delete(struct pipe_context *pipe, void *hwcso)
150 {
151 struct nv30_fragprog *fp = hwcso;
152
153 pipe_resource_reference(&fp->buffer, NULL);
154
155 if (fp->draw)
156 draw_delete_fragment_shader(nv30_context(pipe)->draw, fp->draw);
157
158 FREE((void *)fp->pipe.tokens);
159 FREE(fp->insn);
160 FREE(fp->consts);
161 FREE(fp);
162 }
163
164 static void
nv30_fp_state_bind(struct pipe_context * pipe,void * hwcso)165 nv30_fp_state_bind(struct pipe_context *pipe, void *hwcso)
166 {
167 struct nv30_context *nv30 = nv30_context(pipe);
168 struct nv30_fragprog *fp = hwcso;
169
170 /* reset the bucftx so that we don't keep a dangling reference to the fp
171 * code
172 */
173 if (fp != nv30->state.fragprog)
174 PUSH_RESET(nv30->base.pushbuf, BUFCTX_FRAGPROG);
175
176 nv30->fragprog.program = fp;
177 nv30->dirty |= NV30_NEW_FRAGPROG;
178 }
179
180 void
nv30_fragprog_init(struct pipe_context * pipe)181 nv30_fragprog_init(struct pipe_context *pipe)
182 {
183 pipe->create_fs_state = nv30_fp_state_create;
184 pipe->bind_fs_state = nv30_fp_state_bind;
185 pipe->delete_fs_state = nv30_fp_state_delete;
186 }
187