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 "util/u_inlines.h"
27 #include "pipe/p_state.h"
28 
29 
30 #include "svga_context.h"
31 #include "svga_state.h"
32 #include "svga_debug.h"
33 #include "svga_hw_reg.h"
34 
35 /***********************************************************************
36  */
37 
38 
39 /**
40  * Given a gallium vertex element format, return the corresponding SVGA3D
41  * format.  Return SVGA3D_DECLTYPE_MAX for unsupported gallium formats.
42  */
43 static INLINE SVGA3dDeclType
svga_translate_vertex_format(enum pipe_format format)44 svga_translate_vertex_format(enum pipe_format format)
45 {
46    switch (format) {
47    case PIPE_FORMAT_R32_FLOAT:            return SVGA3D_DECLTYPE_FLOAT1;
48    case PIPE_FORMAT_R32G32_FLOAT:         return SVGA3D_DECLTYPE_FLOAT2;
49    case PIPE_FORMAT_R32G32B32_FLOAT:      return SVGA3D_DECLTYPE_FLOAT3;
50    case PIPE_FORMAT_R32G32B32A32_FLOAT:   return SVGA3D_DECLTYPE_FLOAT4;
51    case PIPE_FORMAT_B8G8R8A8_UNORM:       return SVGA3D_DECLTYPE_D3DCOLOR;
52    case PIPE_FORMAT_R8G8B8A8_USCALED:     return SVGA3D_DECLTYPE_UBYTE4;
53    case PIPE_FORMAT_R16G16_SSCALED:       return SVGA3D_DECLTYPE_SHORT2;
54    case PIPE_FORMAT_R16G16B16A16_SSCALED: return SVGA3D_DECLTYPE_SHORT4;
55    case PIPE_FORMAT_R8G8B8A8_UNORM:       return SVGA3D_DECLTYPE_UBYTE4N;
56    case PIPE_FORMAT_R16G16_SNORM:         return SVGA3D_DECLTYPE_SHORT2N;
57    case PIPE_FORMAT_R16G16B16A16_SNORM:   return SVGA3D_DECLTYPE_SHORT4N;
58    case PIPE_FORMAT_R16G16_UNORM:         return SVGA3D_DECLTYPE_USHORT2N;
59    case PIPE_FORMAT_R16G16B16A16_UNORM:   return SVGA3D_DECLTYPE_USHORT4N;
60    case PIPE_FORMAT_R10G10B10X2_USCALED:  return SVGA3D_DECLTYPE_UDEC3;
61    case PIPE_FORMAT_R10G10B10X2_SNORM:    return SVGA3D_DECLTYPE_DEC3N;
62    case PIPE_FORMAT_R16G16_FLOAT:         return SVGA3D_DECLTYPE_FLOAT16_2;
63    case PIPE_FORMAT_R16G16B16A16_FLOAT:   return SVGA3D_DECLTYPE_FLOAT16_4;
64 
65    default:
66       /* There are many formats without hardware support.  This case
67        * will be hit regularly, meaning we'll need swvfetch.
68        */
69       return SVGA3D_DECLTYPE_MAX;
70    }
71 }
72 
73 
74 static enum pipe_error
update_need_swvfetch(struct svga_context * svga,unsigned dirty)75 update_need_swvfetch( struct svga_context *svga,
76                       unsigned dirty )
77 {
78    unsigned i;
79    boolean need_swvfetch = FALSE;
80 
81    if (!svga->curr.velems) {
82       /* No vertex elements bound. */
83       return 0;
84    }
85 
86    for (i = 0; i < svga->curr.velems->count; i++) {
87       svga->state.sw.ve_format[i] = svga_translate_vertex_format(svga->curr.velems->velem[i].src_format);
88       if (svga->state.sw.ve_format[i] == SVGA3D_DECLTYPE_MAX) {
89          /* Unsupported format - use software fetch */
90          need_swvfetch = TRUE;
91          break;
92       }
93    }
94 
95    if (need_swvfetch != svga->state.sw.need_swvfetch) {
96       svga->state.sw.need_swvfetch = need_swvfetch;
97       svga->dirty |= SVGA_NEW_NEED_SWVFETCH;
98    }
99 
100    return PIPE_OK;
101 }
102 
103 struct svga_tracked_state svga_update_need_swvfetch =
104 {
105    "update need_swvfetch",
106    ( SVGA_NEW_VELEMENT ),
107    update_need_swvfetch
108 };
109 
110 
111 /***********************************************************************
112  */
113 
114 static enum pipe_error
update_need_pipeline(struct svga_context * svga,unsigned dirty)115 update_need_pipeline( struct svga_context *svga,
116                       unsigned dirty )
117 {
118 
119    boolean need_pipeline = FALSE;
120    struct svga_vertex_shader *vs = svga->curr.vs;
121 
122    /* SVGA_NEW_RAST, SVGA_NEW_REDUCED_PRIMITIVE
123     */
124    if (svga->curr.rast->need_pipeline & (1 << svga->curr.reduced_prim)) {
125       SVGA_DBG(DEBUG_SWTNL, "%s: rast need_pipeline (0x%x) & prim (0x%x)\n",
126                  __FUNCTION__,
127                  svga->curr.rast->need_pipeline,
128                  (1 << svga->curr.reduced_prim) );
129       SVGA_DBG(DEBUG_SWTNL, "%s: rast need_pipeline tris (%s), lines (%s), points (%s)\n",
130                  __FUNCTION__,
131                  svga->curr.rast->need_pipeline_tris_str,
132                  svga->curr.rast->need_pipeline_lines_str,
133                  svga->curr.rast->need_pipeline_points_str);
134       need_pipeline = TRUE;
135    }
136 
137    /* EDGEFLAGS
138     */
139     if (vs && vs->base.info.writes_edgeflag) {
140       SVGA_DBG(DEBUG_SWTNL, "%s: edgeflags\n", __FUNCTION__);
141       need_pipeline = TRUE;
142    }
143 
144    /* SVGA_NEW_FS, SVGA_NEW_RAST, SVGA_NEW_REDUCED_PRIMITIVE
145     */
146    if (svga->curr.reduced_prim == PIPE_PRIM_POINTS) {
147       unsigned sprite_coord_gen = svga->curr.rast->templ.sprite_coord_enable;
148       unsigned generic_inputs =
149          svga->curr.fs ? svga->curr.fs->generic_inputs : 0;
150 
151       if (sprite_coord_gen &&
152           (generic_inputs & ~sprite_coord_gen)) {
153          /* The fragment shader is using some generic inputs that are
154           * not being replaced by auto-generated point/sprite coords (and
155           * auto sprite coord generation is turned on).
156           * The SVGA3D interface does not support that: if we enable
157           * SVGA3D_RS_POINTSPRITEENABLE it gets enabled for _all_
158           * texture coordinate sets.
159           * To solve this, we have to use the draw-module's wide/sprite
160           * point stage.
161           */
162          need_pipeline = TRUE;
163       }
164    }
165 
166    if (need_pipeline != svga->state.sw.need_pipeline) {
167       svga->state.sw.need_pipeline = need_pipeline;
168       svga->dirty |= SVGA_NEW_NEED_PIPELINE;
169    }
170 
171    /* DEBUG */
172    if (0 && svga->state.sw.need_pipeline)
173       debug_printf("sw.need_pipeline = %d\n", svga->state.sw.need_pipeline);
174 
175    return PIPE_OK;
176 }
177 
178 
179 struct svga_tracked_state svga_update_need_pipeline =
180 {
181    "need pipeline",
182    (SVGA_NEW_RAST |
183     SVGA_NEW_FS |
184     SVGA_NEW_VS |
185     SVGA_NEW_REDUCED_PRIMITIVE),
186    update_need_pipeline
187 };
188 
189 
190 /***********************************************************************
191  */
192 
193 static enum pipe_error
update_need_swtnl(struct svga_context * svga,unsigned dirty)194 update_need_swtnl( struct svga_context *svga,
195                    unsigned dirty )
196 {
197    boolean need_swtnl;
198 
199    if (svga->debug.no_swtnl) {
200       svga->state.sw.need_swvfetch = FALSE;
201       svga->state.sw.need_pipeline = FALSE;
202    }
203 
204    need_swtnl = (svga->state.sw.need_swvfetch ||
205                  svga->state.sw.need_pipeline);
206 
207    if (svga->debug.force_swtnl) {
208       need_swtnl = TRUE;
209    }
210 
211    /*
212     * Some state changes the draw module does makes us believe we
213     * we don't need swtnl. This causes the vdecl code to pickup
214     * the wrong buffers and vertex formats. Try trivial/line-wide.
215     */
216    if (svga->state.sw.in_swtnl_draw)
217       need_swtnl = TRUE;
218 
219    if (need_swtnl != svga->state.sw.need_swtnl) {
220       SVGA_DBG(DEBUG_SWTNL|DEBUG_PERF,
221                "%s: need_swvfetch %s, need_pipeline %s\n",
222                __FUNCTION__,
223                svga->state.sw.need_swvfetch ? "true" : "false",
224                svga->state.sw.need_pipeline ? "true" : "false");
225 
226       svga->state.sw.need_swtnl = need_swtnl;
227       svga->dirty |= SVGA_NEW_NEED_SWTNL;
228       svga->swtnl.new_vdecl = TRUE;
229    }
230 
231    return PIPE_OK;
232 }
233 
234 
235 struct svga_tracked_state svga_update_need_swtnl =
236 {
237    "need swtnl",
238    (SVGA_NEW_NEED_PIPELINE |
239     SVGA_NEW_NEED_SWVFETCH),
240    update_need_swtnl
241 };
242