1 /**********************************************************
2 * Copyright 2008-2009 VMware, Inc. All rights reserved.
3 *
4 * Permission is hereby granted, free of charge, to any person
5 * obtaining a copy of this software and associated documentation
6 * files (the "Software"), to deal in the Software without
7 * restriction, including without limitation the rights to use, copy,
8 * modify, merge, publish, distribute, sublicense, and/or sell copies
9 * of the Software, and to permit persons to whom the Software is
10 * furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice shall be
13 * included in all copies or substantial portions of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
16 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
17 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
18 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
19 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
20 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
21 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 *
24 **********************************************************/
25
26 #include "pipe/p_defines.h"
27 #include "util/u_bitmask.h"
28 #include "util/u_format.h"
29 #include "util/u_helpers.h"
30 #include "util/u_inlines.h"
31 #include "util/u_math.h"
32 #include "util/u_memory.h"
33 #include "util/u_transfer.h"
34 #include "tgsi/tgsi_parse.h"
35
36 #include "svga_context.h"
37 #include "svga_cmd.h"
38 #include "svga_format.h"
39 #include "svga_resource_buffer.h"
40 #include "svga_screen.h"
41
42
43 static void
svga_set_vertex_buffers(struct pipe_context * pipe,unsigned start_slot,unsigned count,const struct pipe_vertex_buffer * buffers)44 svga_set_vertex_buffers(struct pipe_context *pipe,
45 unsigned start_slot, unsigned count,
46 const struct pipe_vertex_buffer *buffers)
47 {
48 struct svga_context *svga = svga_context(pipe);
49
50 util_set_vertex_buffers_count(svga->curr.vb,
51 &svga->curr.num_vertex_buffers,
52 buffers, start_slot, count);
53
54 svga->dirty |= SVGA_NEW_VBUFFER;
55 }
56
57
58 /**
59 * Does the given vertex attrib format need range adjustment in the VS?
60 * Range adjustment scales and biases values from [0,1] to [-1,1].
61 * This lets us avoid the swtnl path.
62 */
63 static boolean
attrib_needs_range_adjustment(enum pipe_format format)64 attrib_needs_range_adjustment(enum pipe_format format)
65 {
66 switch (format) {
67 case PIPE_FORMAT_R8G8B8_SNORM:
68 return TRUE;
69 default:
70 return FALSE;
71 }
72 }
73
74
75 /**
76 * Given a gallium vertex element format, return the corresponding
77 * SVGA3dDeclType.
78 */
79 static SVGA3dDeclType
translate_vertex_format_to_decltype(enum pipe_format format)80 translate_vertex_format_to_decltype(enum pipe_format format)
81 {
82 switch (format) {
83 case PIPE_FORMAT_R32_FLOAT: return SVGA3D_DECLTYPE_FLOAT1;
84 case PIPE_FORMAT_R32G32_FLOAT: return SVGA3D_DECLTYPE_FLOAT2;
85 case PIPE_FORMAT_R32G32B32_FLOAT: return SVGA3D_DECLTYPE_FLOAT3;
86 case PIPE_FORMAT_R32G32B32A32_FLOAT: return SVGA3D_DECLTYPE_FLOAT4;
87 case PIPE_FORMAT_B8G8R8A8_UNORM: return SVGA3D_DECLTYPE_D3DCOLOR;
88 case PIPE_FORMAT_R8G8B8A8_USCALED: return SVGA3D_DECLTYPE_UBYTE4;
89 case PIPE_FORMAT_R16G16_SSCALED: return SVGA3D_DECLTYPE_SHORT2;
90 case PIPE_FORMAT_R16G16B16A16_SSCALED: return SVGA3D_DECLTYPE_SHORT4;
91 case PIPE_FORMAT_R8G8B8A8_UNORM: return SVGA3D_DECLTYPE_UBYTE4N;
92 case PIPE_FORMAT_R16G16_SNORM: return SVGA3D_DECLTYPE_SHORT2N;
93 case PIPE_FORMAT_R16G16B16A16_SNORM: return SVGA3D_DECLTYPE_SHORT4N;
94 case PIPE_FORMAT_R16G16_UNORM: return SVGA3D_DECLTYPE_USHORT2N;
95 case PIPE_FORMAT_R16G16B16A16_UNORM: return SVGA3D_DECLTYPE_USHORT4N;
96 case PIPE_FORMAT_R10G10B10X2_USCALED: return SVGA3D_DECLTYPE_UDEC3;
97 case PIPE_FORMAT_R10G10B10X2_SNORM: return SVGA3D_DECLTYPE_DEC3N;
98 case PIPE_FORMAT_R16G16_FLOAT: return SVGA3D_DECLTYPE_FLOAT16_2;
99 case PIPE_FORMAT_R16G16B16A16_FLOAT: return SVGA3D_DECLTYPE_FLOAT16_4;
100
101 /* See attrib_needs_adjustment() and attrib_needs_w_to_1() above */
102 case PIPE_FORMAT_R8G8B8_SNORM: return SVGA3D_DECLTYPE_UBYTE4N;
103
104 /* See attrib_needs_w_to_1() above */
105 case PIPE_FORMAT_R16G16B16_SNORM: return SVGA3D_DECLTYPE_SHORT4N;
106 case PIPE_FORMAT_R16G16B16_UNORM: return SVGA3D_DECLTYPE_USHORT4N;
107 case PIPE_FORMAT_R8G8B8_UNORM: return SVGA3D_DECLTYPE_UBYTE4N;
108
109 default:
110 /* There are many formats without hardware support. This case
111 * will be hit regularly, meaning we'll need swvfetch.
112 */
113 return SVGA3D_DECLTYPE_MAX;
114 }
115 }
116
117
118 static void
define_input_element_object(struct svga_context * svga,struct svga_velems_state * velems)119 define_input_element_object(struct svga_context *svga,
120 struct svga_velems_state *velems)
121 {
122 SVGA3dInputElementDesc elements[PIPE_MAX_ATTRIBS];
123 enum pipe_error ret;
124 unsigned i;
125
126 assert(velems->count <= PIPE_MAX_ATTRIBS);
127 assert(svga_have_vgpu10(svga));
128
129 for (i = 0; i < velems->count; i++) {
130 const struct pipe_vertex_element *elem = velems->velem + i;
131 SVGA3dSurfaceFormat svga_format;
132 unsigned vf_flags;
133
134 svga_translate_vertex_format_vgpu10(elem->src_format,
135 &svga_format, &vf_flags);
136
137 velems->decl_type[i] =
138 translate_vertex_format_to_decltype(elem->src_format);
139 elements[i].inputSlot = elem->vertex_buffer_index;
140 elements[i].alignedByteOffset = elem->src_offset;
141 elements[i].format = svga_format;
142
143 if (elem->instance_divisor) {
144 elements[i].inputSlotClass = SVGA3D_INPUT_PER_INSTANCE_DATA;
145 elements[i].instanceDataStepRate = elem->instance_divisor;
146 }
147 else {
148 elements[i].inputSlotClass = SVGA3D_INPUT_PER_VERTEX_DATA;
149 elements[i].instanceDataStepRate = 0;
150 }
151 elements[i].inputRegister = i;
152
153 if (elements[i].format == SVGA3D_FORMAT_INVALID) {
154 velems->need_swvfetch = TRUE;
155 }
156
157 if (util_format_is_pure_integer(elem->src_format)) {
158 velems->attrib_is_pure_int |= (1 << i);
159 }
160
161 if (vf_flags & VF_W_TO_1) {
162 velems->adjust_attrib_w_1 |= (1 << i);
163 }
164
165 if (vf_flags & VF_U_TO_F_CAST) {
166 velems->adjust_attrib_utof |= (1 << i);
167 }
168 else if (vf_flags & VF_I_TO_F_CAST) {
169 velems->adjust_attrib_itof |= (1 << i);
170 }
171
172 if (vf_flags & VF_BGRA) {
173 velems->attrib_is_bgra |= (1 << i);
174 }
175
176 if (vf_flags & VF_PUINT_TO_SNORM) {
177 velems->attrib_puint_to_snorm |= (1 << i);
178 }
179 else if (vf_flags & VF_PUINT_TO_USCALED) {
180 velems->attrib_puint_to_uscaled |= (1 << i);
181 }
182 else if (vf_flags & VF_PUINT_TO_SSCALED) {
183 velems->attrib_puint_to_sscaled |= (1 << i);
184 }
185 }
186
187 velems->id = util_bitmask_add(svga->input_element_object_id_bm);
188
189 ret = SVGA3D_vgpu10_DefineElementLayout(svga->swc, velems->count,
190 velems->id, elements);
191 if (ret != PIPE_OK) {
192 svga_context_flush(svga, NULL);
193 ret = SVGA3D_vgpu10_DefineElementLayout(svga->swc, velems->count,
194 velems->id, elements);
195 assert(ret == PIPE_OK);
196 }
197 }
198
199
200 /**
201 * Translate the vertex element types to SVGA3dDeclType and check
202 * for VS-based vertex attribute adjustments.
203 */
204 static void
translate_vertex_decls(struct svga_context * svga,struct svga_velems_state * velems)205 translate_vertex_decls(struct svga_context *svga,
206 struct svga_velems_state *velems)
207 {
208 unsigned i;
209
210 assert(!svga_have_vgpu10(svga));
211
212 for (i = 0; i < velems->count; i++) {
213 const enum pipe_format f = velems->velem[i].src_format;
214 SVGA3dSurfaceFormat svga_format;
215 unsigned vf_flags;
216
217 svga_translate_vertex_format_vgpu10(f, &svga_format, &vf_flags);
218
219 velems->decl_type[i] = translate_vertex_format_to_decltype(f);
220 if (velems->decl_type[i] == SVGA3D_DECLTYPE_MAX) {
221 /* Unsupported format - use software fetch */
222 velems->need_swvfetch = TRUE;
223 }
224
225 /* Check for VS-based adjustments */
226 if (attrib_needs_range_adjustment(f)) {
227 velems->adjust_attrib_range |= (1 << i);
228 }
229
230 if (vf_flags & VF_W_TO_1) {
231 velems->adjust_attrib_w_1 |= (1 << i);
232 }
233 }
234 }
235
236
237 static void *
svga_create_vertex_elements_state(struct pipe_context * pipe,unsigned count,const struct pipe_vertex_element * attribs)238 svga_create_vertex_elements_state(struct pipe_context *pipe,
239 unsigned count,
240 const struct pipe_vertex_element *attribs)
241 {
242 struct svga_context *svga = svga_context(pipe);
243 struct svga_velems_state *velems;
244
245 assert(count <= PIPE_MAX_ATTRIBS);
246 velems = (struct svga_velems_state *) MALLOC(sizeof(struct svga_velems_state));
247 if (velems) {
248 velems->count = count;
249 memcpy(velems->velem, attribs, sizeof(*attribs) * count);
250
251 velems->need_swvfetch = FALSE;
252 velems->adjust_attrib_range = 0x0;
253 velems->attrib_is_pure_int = 0x0;
254 velems->adjust_attrib_w_1 = 0x0;
255 velems->adjust_attrib_itof = 0x0;
256 velems->adjust_attrib_utof = 0x0;
257 velems->attrib_is_bgra = 0x0;
258 velems->attrib_puint_to_snorm = 0x0;
259 velems->attrib_puint_to_uscaled = 0x0;
260 velems->attrib_puint_to_sscaled = 0x0;
261
262 if (svga_have_vgpu10(svga)) {
263 define_input_element_object(svga, velems);
264 }
265 else {
266 translate_vertex_decls(svga, velems);
267 }
268 }
269
270 svga->hud.num_vertexelement_objects++;
271 SVGA_STATS_COUNT_INC(svga_screen(svga->pipe.screen)->sws,
272 SVGA_STATS_COUNT_VERTEXELEMENT);
273
274 return velems;
275 }
276
277
278 static void
svga_bind_vertex_elements_state(struct pipe_context * pipe,void * state)279 svga_bind_vertex_elements_state(struct pipe_context *pipe, void *state)
280 {
281 struct svga_context *svga = svga_context(pipe);
282 struct svga_velems_state *velems = (struct svga_velems_state *) state;
283
284 svga->curr.velems = velems;
285 svga->dirty |= SVGA_NEW_VELEMENT;
286 }
287
288
289 static void
svga_delete_vertex_elements_state(struct pipe_context * pipe,void * state)290 svga_delete_vertex_elements_state(struct pipe_context *pipe, void *state)
291 {
292 struct svga_context *svga = svga_context(pipe);
293 struct svga_velems_state *velems = (struct svga_velems_state *) state;
294
295 if (svga_have_vgpu10(svga)) {
296 enum pipe_error ret;
297
298 svga_hwtnl_flush_retry(svga);
299
300 ret = SVGA3D_vgpu10_DestroyElementLayout(svga->swc, velems->id);
301 if (ret != PIPE_OK) {
302 svga_context_flush(svga, NULL);
303 ret = SVGA3D_vgpu10_DestroyElementLayout(svga->swc, velems->id);
304 assert(ret == PIPE_OK);
305 }
306
307 if (velems->id == svga->state.hw_draw.layout_id)
308 svga->state.hw_draw.layout_id = SVGA3D_INVALID_ID;
309
310 util_bitmask_clear(svga->input_element_object_id_bm, velems->id);
311 velems->id = SVGA3D_INVALID_ID;
312 }
313
314 FREE(velems);
315 svga->hud.num_vertexelement_objects--;
316 }
317
318
319 void
svga_cleanup_vertex_state(struct svga_context * svga)320 svga_cleanup_vertex_state(struct svga_context *svga)
321 {
322 unsigned i;
323
324 for (i = 0 ; i < svga->curr.num_vertex_buffers; i++)
325 pipe_vertex_buffer_unreference(&svga->curr.vb[i]);
326
327 pipe_resource_reference(&svga->state.hw_draw.ib, NULL);
328
329 for (i = 0; i < svga->state.hw_draw.num_vbuffers; i++)
330 pipe_resource_reference(&svga->state.hw_draw.vbuffers[i], NULL);
331 }
332
333
334 void
svga_init_vertex_functions(struct svga_context * svga)335 svga_init_vertex_functions(struct svga_context *svga)
336 {
337 svga->pipe.set_vertex_buffers = svga_set_vertex_buffers;
338 svga->pipe.create_vertex_elements_state = svga_create_vertex_elements_state;
339 svga->pipe.bind_vertex_elements_state = svga_bind_vertex_elements_state;
340 svga->pipe.delete_vertex_elements_state = svga_delete_vertex_elements_state;
341 }
342