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
svga_set_vertex_buffers(struct pipe_context * pipe,unsigned start_slot,unsigned count,const struct pipe_vertex_buffer * buffers)43 static void svga_set_vertex_buffers(struct pipe_context *pipe,
44 unsigned start_slot, unsigned count,
45 const struct pipe_vertex_buffer *buffers)
46 {
47 struct svga_context *svga = svga_context(pipe);
48
49 util_set_vertex_buffers_count(svga->curr.vb,
50 &svga->curr.num_vertex_buffers,
51 buffers, start_slot, count);
52
53 svga->dirty |= SVGA_NEW_VBUFFER;
54 }
55
56
svga_set_index_buffer(struct pipe_context * pipe,const struct pipe_index_buffer * ib)57 static void svga_set_index_buffer(struct pipe_context *pipe,
58 const struct pipe_index_buffer *ib)
59 {
60 struct svga_context *svga = svga_context(pipe);
61
62 util_set_index_buffer(&svga->curr.ib, ib);
63 }
64
65
66 /**
67 * Does the given vertex attrib format need range adjustment in the VS?
68 * Range adjustment scales and biases values from [0,1] to [-1,1].
69 * This lets us avoid the swtnl path.
70 */
71 static boolean
attrib_needs_range_adjustment(enum pipe_format format)72 attrib_needs_range_adjustment(enum pipe_format format)
73 {
74 switch (format) {
75 case PIPE_FORMAT_R8G8B8_SNORM:
76 return TRUE;
77 default:
78 return FALSE;
79 }
80 }
81
82
83 /**
84 * Given a gallium vertex element format, return the corresponding
85 * SVGA3dDeclType.
86 */
87 static SVGA3dDeclType
translate_vertex_format_to_decltype(enum pipe_format format)88 translate_vertex_format_to_decltype(enum pipe_format format)
89 {
90 switch (format) {
91 case PIPE_FORMAT_R32_FLOAT: return SVGA3D_DECLTYPE_FLOAT1;
92 case PIPE_FORMAT_R32G32_FLOAT: return SVGA3D_DECLTYPE_FLOAT2;
93 case PIPE_FORMAT_R32G32B32_FLOAT: return SVGA3D_DECLTYPE_FLOAT3;
94 case PIPE_FORMAT_R32G32B32A32_FLOAT: return SVGA3D_DECLTYPE_FLOAT4;
95 case PIPE_FORMAT_B8G8R8A8_UNORM: return SVGA3D_DECLTYPE_D3DCOLOR;
96 case PIPE_FORMAT_R8G8B8A8_USCALED: return SVGA3D_DECLTYPE_UBYTE4;
97 case PIPE_FORMAT_R16G16_SSCALED: return SVGA3D_DECLTYPE_SHORT2;
98 case PIPE_FORMAT_R16G16B16A16_SSCALED: return SVGA3D_DECLTYPE_SHORT4;
99 case PIPE_FORMAT_R8G8B8A8_UNORM: return SVGA3D_DECLTYPE_UBYTE4N;
100 case PIPE_FORMAT_R16G16_SNORM: return SVGA3D_DECLTYPE_SHORT2N;
101 case PIPE_FORMAT_R16G16B16A16_SNORM: return SVGA3D_DECLTYPE_SHORT4N;
102 case PIPE_FORMAT_R16G16_UNORM: return SVGA3D_DECLTYPE_USHORT2N;
103 case PIPE_FORMAT_R16G16B16A16_UNORM: return SVGA3D_DECLTYPE_USHORT4N;
104 case PIPE_FORMAT_R10G10B10X2_USCALED: return SVGA3D_DECLTYPE_UDEC3;
105 case PIPE_FORMAT_R10G10B10X2_SNORM: return SVGA3D_DECLTYPE_DEC3N;
106 case PIPE_FORMAT_R16G16_FLOAT: return SVGA3D_DECLTYPE_FLOAT16_2;
107 case PIPE_FORMAT_R16G16B16A16_FLOAT: return SVGA3D_DECLTYPE_FLOAT16_4;
108
109 /* See attrib_needs_adjustment() and attrib_needs_w_to_1() above */
110 case PIPE_FORMAT_R8G8B8_SNORM: return SVGA3D_DECLTYPE_UBYTE4N;
111
112 /* See attrib_needs_w_to_1() above */
113 case PIPE_FORMAT_R16G16B16_SNORM: return SVGA3D_DECLTYPE_SHORT4N;
114 case PIPE_FORMAT_R16G16B16_UNORM: return SVGA3D_DECLTYPE_USHORT4N;
115 case PIPE_FORMAT_R8G8B8_UNORM: return SVGA3D_DECLTYPE_UBYTE4N;
116
117 default:
118 /* There are many formats without hardware support. This case
119 * will be hit regularly, meaning we'll need swvfetch.
120 */
121 return SVGA3D_DECLTYPE_MAX;
122 }
123 }
124
125
126 static void
define_input_element_object(struct svga_context * svga,struct svga_velems_state * velems)127 define_input_element_object(struct svga_context *svga,
128 struct svga_velems_state *velems)
129 {
130 SVGA3dInputElementDesc elements[PIPE_MAX_ATTRIBS];
131 enum pipe_error ret;
132 unsigned i;
133
134 assert(velems->count <= PIPE_MAX_ATTRIBS);
135 assert(svga_have_vgpu10(svga));
136
137 for (i = 0; i < velems->count; i++) {
138 const struct pipe_vertex_element *elem = velems->velem + i;
139 SVGA3dSurfaceFormat svga_format;
140 unsigned vf_flags;
141
142 svga_translate_vertex_format_vgpu10(elem->src_format,
143 &svga_format, &vf_flags);
144
145 velems->decl_type[i] =
146 translate_vertex_format_to_decltype(elem->src_format);
147 elements[i].inputSlot = elem->vertex_buffer_index;
148 elements[i].alignedByteOffset = elem->src_offset;
149 elements[i].format = svga_format;
150
151 if (elem->instance_divisor) {
152 elements[i].inputSlotClass = SVGA3D_INPUT_PER_INSTANCE_DATA;
153 elements[i].instanceDataStepRate = elem->instance_divisor;
154 }
155 else {
156 elements[i].inputSlotClass = SVGA3D_INPUT_PER_VERTEX_DATA;
157 elements[i].instanceDataStepRate = 0;
158 }
159 elements[i].inputRegister = i;
160
161 if (elements[i].format == SVGA3D_FORMAT_INVALID) {
162 velems->need_swvfetch = TRUE;
163 }
164
165 if (util_format_is_pure_integer(elem->src_format)) {
166 velems->attrib_is_pure_int |= (1 << i);
167 }
168
169 if (vf_flags & VF_W_TO_1) {
170 velems->adjust_attrib_w_1 |= (1 << i);
171 }
172
173 if (vf_flags & VF_U_TO_F_CAST) {
174 velems->adjust_attrib_utof |= (1 << i);
175 }
176 else if (vf_flags & VF_I_TO_F_CAST) {
177 velems->adjust_attrib_itof |= (1 << i);
178 }
179
180 if (vf_flags & VF_BGRA) {
181 velems->attrib_is_bgra |= (1 << i);
182 }
183
184 if (vf_flags & VF_PUINT_TO_SNORM) {
185 velems->attrib_puint_to_snorm |= (1 << i);
186 }
187 else if (vf_flags & VF_PUINT_TO_USCALED) {
188 velems->attrib_puint_to_uscaled |= (1 << i);
189 }
190 else if (vf_flags & VF_PUINT_TO_SSCALED) {
191 velems->attrib_puint_to_sscaled |= (1 << i);
192 }
193 }
194
195 velems->id = util_bitmask_add(svga->input_element_object_id_bm);
196
197 ret = SVGA3D_vgpu10_DefineElementLayout(svga->swc, velems->count,
198 velems->id, elements);
199 if (ret != PIPE_OK) {
200 svga_context_flush(svga, NULL);
201 ret = SVGA3D_vgpu10_DefineElementLayout(svga->swc, velems->count,
202 velems->id, elements);
203 assert(ret == PIPE_OK);
204 }
205 }
206
207
208 /**
209 * Translate the vertex element types to SVGA3dDeclType and check
210 * for VS-based vertex attribute adjustments.
211 */
212 static void
translate_vertex_decls(struct svga_context * svga,struct svga_velems_state * velems)213 translate_vertex_decls(struct svga_context *svga,
214 struct svga_velems_state *velems)
215 {
216 unsigned i;
217
218 assert(!svga_have_vgpu10(svga));
219
220 for (i = 0; i < velems->count; i++) {
221 const enum pipe_format f = velems->velem[i].src_format;
222 SVGA3dSurfaceFormat svga_format;
223 unsigned vf_flags;
224
225 svga_translate_vertex_format_vgpu10(f, &svga_format, &vf_flags);
226
227 velems->decl_type[i] = translate_vertex_format_to_decltype(f);
228 if (velems->decl_type[i] == SVGA3D_DECLTYPE_MAX) {
229 /* Unsupported format - use software fetch */
230 velems->need_swvfetch = TRUE;
231 }
232
233 /* Check for VS-based adjustments */
234 if (attrib_needs_range_adjustment(f)) {
235 velems->adjust_attrib_range |= (1 << i);
236 }
237
238 if (vf_flags & VF_W_TO_1) {
239 velems->adjust_attrib_w_1 |= (1 << i);
240 }
241 }
242 }
243
244
245 static void *
svga_create_vertex_elements_state(struct pipe_context * pipe,unsigned count,const struct pipe_vertex_element * attribs)246 svga_create_vertex_elements_state(struct pipe_context *pipe,
247 unsigned count,
248 const struct pipe_vertex_element *attribs)
249 {
250 struct svga_context *svga = svga_context(pipe);
251 struct svga_velems_state *velems;
252
253 assert(count <= PIPE_MAX_ATTRIBS);
254 velems = (struct svga_velems_state *) MALLOC(sizeof(struct svga_velems_state));
255 if (velems) {
256 velems->count = count;
257 memcpy(velems->velem, attribs, sizeof(*attribs) * count);
258
259 velems->need_swvfetch = FALSE;
260 velems->adjust_attrib_range = 0x0;
261 velems->attrib_is_pure_int = 0x0;
262 velems->adjust_attrib_w_1 = 0x0;
263 velems->adjust_attrib_itof = 0x0;
264 velems->adjust_attrib_utof = 0x0;
265 velems->attrib_is_bgra = 0x0;
266 velems->attrib_puint_to_snorm = 0x0;
267 velems->attrib_puint_to_uscaled = 0x0;
268 velems->attrib_puint_to_sscaled = 0x0;
269
270 if (svga_have_vgpu10(svga)) {
271 define_input_element_object(svga, velems);
272 }
273 else {
274 translate_vertex_decls(svga, velems);
275 }
276 }
277
278 svga->hud.num_vertexelement_objects++;
279 SVGA_STATS_COUNT_INC(svga_screen(svga->pipe.screen)->sws,
280 SVGA_STATS_COUNT_VERTEXELEMENT);
281
282 return velems;
283 }
284
285
286 static void
svga_bind_vertex_elements_state(struct pipe_context * pipe,void * state)287 svga_bind_vertex_elements_state(struct pipe_context *pipe, void *state)
288 {
289 struct svga_context *svga = svga_context(pipe);
290 struct svga_velems_state *velems = (struct svga_velems_state *) state;
291
292 svga->curr.velems = velems;
293 svga->dirty |= SVGA_NEW_VELEMENT;
294 }
295
296
297 static void
svga_delete_vertex_elements_state(struct pipe_context * pipe,void * state)298 svga_delete_vertex_elements_state(struct pipe_context *pipe, void *state)
299 {
300 struct svga_context *svga = svga_context(pipe);
301 struct svga_velems_state *velems = (struct svga_velems_state *) state;
302
303 if (svga_have_vgpu10(svga)) {
304 enum pipe_error ret;
305
306 svga_hwtnl_flush_retry(svga);
307
308 ret = SVGA3D_vgpu10_DestroyElementLayout(svga->swc, velems->id);
309 if (ret != PIPE_OK) {
310 svga_context_flush(svga, NULL);
311 ret = SVGA3D_vgpu10_DestroyElementLayout(svga->swc, velems->id);
312 assert(ret == PIPE_OK);
313 }
314
315 if (velems->id == svga->state.hw_draw.layout_id)
316 svga->state.hw_draw.layout_id = SVGA3D_INVALID_ID;
317
318 util_bitmask_clear(svga->input_element_object_id_bm, velems->id);
319 velems->id = SVGA3D_INVALID_ID;
320 }
321
322 FREE(velems);
323 svga->hud.num_vertexelement_objects--;
324 }
325
svga_cleanup_vertex_state(struct svga_context * svga)326 void svga_cleanup_vertex_state( struct svga_context *svga )
327 {
328 unsigned i;
329
330 for (i = 0 ; i < svga->curr.num_vertex_buffers; i++)
331 pipe_resource_reference(&svga->curr.vb[i].buffer, NULL);
332
333 pipe_resource_reference(&svga->state.hw_draw.ib, NULL);
334
335 for (i = 0; i < svga->state.hw_draw.num_vbuffers; i++)
336 pipe_resource_reference(&svga->state.hw_draw.vbuffers[i], NULL);
337 }
338
339
svga_init_vertex_functions(struct svga_context * svga)340 void svga_init_vertex_functions( struct svga_context *svga )
341 {
342 svga->pipe.set_vertex_buffers = svga_set_vertex_buffers;
343 svga->pipe.set_index_buffer = svga_set_index_buffer;
344 svga->pipe.create_vertex_elements_state = svga_create_vertex_elements_state;
345 svga->pipe.bind_vertex_elements_state = svga_bind_vertex_elements_state;
346 svga->pipe.delete_vertex_elements_state = svga_delete_vertex_elements_state;
347 }
348