1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
|
#include "bufferobj.h"
#include "enums.h"
#include "nouveau_bufferobj.h"
#include "nouveau_buffers.h"
#include "nouveau_context.h"
#include "nouveau_drm.h"
#include "nouveau_object.h"
#include "nouveau_msg.h"
#define DEBUG(fmt,args...) do { \
if (NOUVEAU_DEBUG & DEBUG_BUFFEROBJ) { \
fprintf(stderr, "%s: "fmt, __func__, ##args); \
} \
} while(0)
/* Wrapper for nouveau_mem_gpu_offset_get() that marks the bufferobj dirty
* if the GPU modifies the data.
*/
uint32_t
nouveau_bufferobj_gpu_ref(GLcontext *ctx, GLenum access,
struct gl_buffer_object *obj)
{
nouveau_buffer_object *nbo = (nouveau_buffer_object *)obj;
DEBUG("obj=%p, access=%s\n", obj, _mesa_lookup_enum_by_nr(access));
if (access == GL_WRITE_ONLY_ARB || access == GL_READ_WRITE_ARB)
nbo->gpu_dirty = GL_TRUE;
return nouveau_mem_gpu_offset_get(ctx, nbo->gpu_mem);
}
static void
nouveauBindBuffer(GLcontext *ctx, GLenum target, struct gl_buffer_object *obj)
{
}
static struct gl_buffer_object *
nouveauNewBufferObject(GLcontext *ctx, GLuint buffer, GLenum target)
{
nouveau_buffer_object *nbo;
nbo = CALLOC_STRUCT(nouveau_buffer_object_t);
DEBUG("name=0x%08x, target=%s, obj=%p\n",
buffer, _mesa_lookup_enum_by_nr(target), nbo);
_mesa_initialize_buffer_object(&nbo->mesa, buffer, target);
return &nbo->mesa;
}
static void
nouveauDeleteBuffer(GLcontext *ctx, struct gl_buffer_object *obj)
{
nouveau_buffer_object *nbo = (nouveau_buffer_object *)obj;
DEBUG("obj=%p\n", obj);
if (nbo->gpu_mem) {
nouveau_mem_free(ctx, nbo->gpu_mem);
}
_mesa_delete_buffer_object(ctx, obj);
}
static void
nouveauBufferData(GLcontext *ctx, GLenum target, GLsizeiptrARB size,
const GLvoid *data, GLenum usage,
struct gl_buffer_object *obj)
{
nouveau_buffer_object *nbo = (nouveau_buffer_object *)obj;
DEBUG("obj=%p, target=%s, usage=%s, size=%d, data=%p\n",
obj,
_mesa_lookup_enum_by_nr(target),
_mesa_lookup_enum_by_nr(usage),
(unsigned int)size,
data);
if (nbo->gpu_mem && nbo->gpu_mem->size != size)
nouveau_mem_free(ctx, nbo->gpu_mem);
/* Always have the GPU access the data from VRAM if possible. For
* some "usage" values it may be better from AGP be default?
*
* TODO: At some point we should drop the NOUVEAU_MEM_MAPPED flag.
* TODO: Use the NOUVEAU_MEM_AGP_ACCEPTABLE flag.
* TODO: What about PCI-E and shared system memory?
*/
if (!nbo->gpu_mem)
nbo->gpu_mem = nouveau_mem_alloc(ctx,
NOUVEAU_MEM_FB |
NOUVEAU_MEM_MAPPED,
size,
0);
if (!nbo->gpu_mem) {
MESSAGE("AIII bufferobj malloc failed\n");
return;
}
obj->Usage = usage;
obj->Size = size;
if (!data)
return;
ctx->Driver.MapBuffer(ctx, target, GL_WRITE_ONLY_ARB, obj);
_mesa_memcpy(nbo->cpu_mem->map, data, size);
ctx->Driver.UnmapBuffer(ctx, target, obj);
}
/*TODO: we don't need to DMA the entire buffer like MapBuffer does.. */
static void
nouveauBufferSubData(GLcontext *ctx, GLenum target, GLintptrARB offset,
GLsizeiptrARB size, const GLvoid *data,
struct gl_buffer_object *obj)
{
DEBUG("obj=%p, target=%s, offset=0x%x, size=%d, data=%p\n",
obj,
_mesa_lookup_enum_by_nr(target),
(unsigned int)offset,
(unsigned int)size,
data);
ctx->Driver.MapBuffer(ctx, target, GL_WRITE_ONLY_ARB, obj);
_mesa_memcpy((GLubyte *)obj->Pointer + offset, data, size);
ctx->Driver.UnmapBuffer(ctx, target, obj);
}
/*TODO: we don't need to DMA the entire buffer like MapBuffer does.. */
static void
nouveauGetBufferSubData(GLcontext *ctx, GLenum target, GLintptrARB offset,
GLsizeiptrARB size, GLvoid *data,
struct gl_buffer_object *obj)
{
DEBUG("obj=%p, target=%s, offset=0x%x, size=%d, data=%p\n",
obj,
_mesa_lookup_enum_by_nr(target),
(unsigned int)offset,
(unsigned int)size,
data);
ctx->Driver.MapBuffer(ctx, target, GL_READ_ONLY_ARB, obj);
_mesa_memcpy(data, (GLubyte *)obj->Pointer + offset, size);
ctx->Driver.UnmapBuffer(ctx, target, obj);
}
static void *
nouveauMapBuffer(GLcontext *ctx, GLenum target, GLenum access,
struct gl_buffer_object *obj)
{
nouveauContextPtr nmesa = NOUVEAU_CONTEXT(ctx);
nouveau_buffer_object *nbo = (nouveau_buffer_object *)obj;
DEBUG("obj=%p, target=%s, access=%s\n",
obj,
_mesa_lookup_enum_by_nr(target),
_mesa_lookup_enum_by_nr(access));
if (obj->Pointer) {
DEBUG("already mapped, return NULL\n");
return NULL;
}
#ifdef ALLOW_MULTI_SUBCHANNEL
/* If GPU is accessing the data from VRAM, copy to faster AGP memory
* before CPU access to the buffer.
*/
if (nbo->gpu_mem->type & NOUVEAU_MEM_FB) {
DEBUG("Data in VRAM, copying to AGP for CPU access\n");
/* This can happen if BufferData grows the GPU-access buffer */
if (nbo->cpu_mem && nbo->cpu_mem->size != nbo->gpu_mem->size) {
nouveau_mem_free(ctx, nbo->cpu_mem);
nbo->cpu_mem = NULL;
}
if (!nbo->cpu_mem) {
nbo->cpu_mem = nouveau_mem_alloc(ctx,
NOUVEAU_MEM_AGP |
NOUVEAU_MEM_MAPPED,
nbo->gpu_mem->size,
0);
/* Mark GPU data as modified, so it gets copied to
* the new buffer */
nbo->gpu_dirty = GL_TRUE;
}
if (nbo->cpu_mem && nbo->gpu_dirty) {
nouveau_memformat_flat_emit(ctx, nbo->cpu_mem,
nbo->gpu_mem,
0, 0,
nbo->gpu_mem->size);
nouveau_notifier_wait_nop(ctx,
nmesa->syncNotifier,
NvSubMemFormat);
nbo->gpu_dirty = GL_FALSE;
}
/* buffer isn't guaranteed to be up-to-date on the card now */
nbo->cpu_dirty = GL_TRUE;
}
#endif
/* If the copy to AGP failed for some reason, just return a pointer
* directly to vram..
*/
if (!nbo->cpu_mem) {
DEBUG("Returning direct pointer to VRAM\n");
nbo->cpu_mem = nbo->gpu_mem;
nbo->cpu_dirty = GL_FALSE;
}
obj->Pointer = nbo->cpu_mem->map;
return obj->Pointer;
}
static GLboolean
nouveauUnmapBuffer(GLcontext *ctx, GLenum target, struct gl_buffer_object *obj)
{
nouveauContextPtr nmesa = NOUVEAU_CONTEXT(ctx);
nouveau_buffer_object *nbo = (nouveau_buffer_object *)obj;
DEBUG("obj=%p, target=%s\n", obj, _mesa_lookup_enum_by_nr(target));
#ifdef ALLOW_MULTI_SUBCHANNEL
if (nbo->cpu_dirty && nbo->cpu_mem != nbo->gpu_mem) {
DEBUG("Copying potentially modified data back to GPU\n");
/* blit from GPU buffer -> CPU buffer */
nouveau_memformat_flat_emit(ctx, nbo->gpu_mem, nbo->cpu_mem,
0, 0, nbo->cpu_mem->size);
/* buffer is now up-to-date on the hardware (or rather, will
* be by the time any other commands in this channel reference
* the data.)
*/
nbo->cpu_dirty = GL_FALSE;
/* we can avoid this wait in some cases.. */
nouveau_notifier_wait_nop(ctx,
nmesa->syncNotifier,
NvSubMemFormat);
/* If it's likely CPU access to the buffer will occur often,
* keep the cpu_mem around to avoid repeated allocs.
*/
if (obj->Usage != GL_DYNAMIC_DRAW_ARB) {
nouveau_mem_free(ctx, nbo->cpu_mem);
nbo->cpu_mem = NULL;
}
}
#endif
obj->Pointer = NULL;
return GL_TRUE;
}
void
nouveauInitBufferObjects(GLcontext *ctx)
{
ctx->Driver.BindBuffer = nouveauBindBuffer;
ctx->Driver.NewBufferObject = nouveauNewBufferObject;
ctx->Driver.DeleteBuffer = nouveauDeleteBuffer;
ctx->Driver.BufferData = nouveauBufferData;
ctx->Driver.BufferSubData = nouveauBufferSubData;
ctx->Driver.GetBufferSubData = nouveauGetBufferSubData;
ctx->Driver.MapBuffer = nouveauMapBuffer;
ctx->Driver.UnmapBuffer = nouveauUnmapBuffer;
}
|
