• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /**************************************************************************
2  *
3  * Copyright 2003 VMware, Inc.
4  * All Rights Reserved.
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a
7  * copy of this software and associated documentation files (the
8  * "Software"), to deal in the Software without restriction, including
9  * without limitation the rights to use, copy, modify, merge, publish,
10  * distribute, sub license, and/or sell copies of the Software, and to
11  * permit persons to whom the Software is furnished to do so, subject to
12  * the following conditions:
13  *
14  * The above copyright notice and this permission notice (including the
15  * next paragraph) shall be included in all copies or substantial portions
16  * of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21  * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
25  *
26  **************************************************************************/
27 
28 
29 #include <stdio.h>
30 #include "main/arrayobj.h"
31 #include "main/glheader.h"
32 #include "main/context.h"
33 
34 #include "pipe/p_defines.h"
35 #include "st_context.h"
36 #include "st_atom.h"
37 #include "st_program.h"
38 #include "st_manager.h"
39 #include "st_util.h"
40 
41 #include "util/u_cpu_detect.h"
42 
43 
44 typedef void (*update_func_t)(struct st_context *st);
45 
46 /* The list state update functions. */
47 static update_func_t update_functions[ST_NUM_ATOMS];
48 
49 static void
init_atoms_once(void)50 init_atoms_once(void)
51 {
52 #define ST_STATE(FLAG, st_update) update_functions[FLAG##_INDEX] = st_update;
53 #include "st_atom_list.h"
54 #undef ST_STATE
55 
56    if (util_get_cpu_caps()->has_popcnt)
57       update_functions[ST_NEW_VERTEX_ARRAYS_INDEX] = st_update_array_with_popcnt;
58 }
59 
st_init_atoms(struct st_context * st)60 void st_init_atoms( struct st_context *st )
61 {
62    STATIC_ASSERT(ARRAY_SIZE(update_functions) <= 64);
63 
64    static once_flag flag = ONCE_FLAG_INIT;
65    call_once(&flag, init_atoms_once);
66 }
67 
68 
st_destroy_atoms(struct st_context * st)69 void st_destroy_atoms( struct st_context *st )
70 {
71    /* no-op */
72 }
73 
74 
75 /* Too complex to figure out, just check every time:
76  */
check_program_state(struct st_context * st)77 static void check_program_state( struct st_context *st )
78 {
79    struct gl_context *ctx = st->ctx;
80    struct gl_program *old_vp = st->vp;
81    struct gl_program *old_tcp = st->tcp;
82    struct gl_program *old_tep = st->tep;
83    struct gl_program *old_gp = st->gp;
84    struct gl_program *old_fp = st->fp;
85 
86    struct gl_program *new_vp = ctx->VertexProgram._Current;
87    struct gl_program *new_tcp = ctx->TessCtrlProgram._Current;
88    struct gl_program *new_tep = ctx->TessEvalProgram._Current;
89    struct gl_program *new_gp = ctx->GeometryProgram._Current;
90    struct gl_program *new_fp = ctx->FragmentProgram._Current;
91    uint64_t dirty = 0;
92    unsigned num_viewports = 1;
93 
94    /* Flag states used by both new and old shaders to unbind shader resources
95     * properly when transitioning to shaders that don't use them.
96     */
97    if (unlikely(new_vp != old_vp)) {
98       ctx->Array.NewVertexElements = true;
99       if (old_vp)
100          dirty |= old_vp->affected_states;
101       if (new_vp)
102          dirty |= ST_NEW_VERTEX_PROGRAM(st, new_vp);
103    }
104 
105    if (unlikely(new_tcp != old_tcp)) {
106       if (old_tcp)
107          dirty |= old_tcp->affected_states;
108       if (new_tcp)
109          dirty |= new_tcp->affected_states;
110    }
111 
112    if (unlikely(new_tep != old_tep)) {
113       if (old_tep)
114          dirty |= old_tep->affected_states;
115       if (new_tep)
116          dirty |= new_tep->affected_states;
117    }
118 
119    if (unlikely(new_gp != old_gp)) {
120       if (old_gp)
121          dirty |= old_gp->affected_states;
122       if (new_gp)
123          dirty |= new_gp->affected_states;
124    }
125 
126    if (unlikely(new_fp != old_fp)) {
127       if (old_fp)
128          dirty |= old_fp->affected_states;
129       if (new_fp)
130          dirty |= new_fp->affected_states;
131    }
132 
133    /* Find out the number of viewports. This determines how many scissors
134     * and viewport states we need to update.
135     */
136    struct gl_program *last_prim_shader = new_gp ? new_gp :
137                                          new_tep ? new_tep : new_vp;
138    if (last_prim_shader &&
139        last_prim_shader->info.outputs_written & (
140              VARYING_BIT_VIEWPORT | VARYING_BIT_VIEWPORT_MASK))
141       num_viewports = ctx->Const.MaxViewports;
142 
143    if (st->state.num_viewports != num_viewports) {
144       st->state.num_viewports = num_viewports;
145       dirty |= ST_NEW_VIEWPORT;
146 
147       if (ctx->Scissor.EnableFlags & u_bit_consecutive(0, num_viewports))
148          dirty |= ST_NEW_SCISSOR;
149    }
150 
151    if (st->lower_point_size && st->ctx->LastVertexStageDirty &&
152        !st->ctx->VertexProgram.PointSizeEnabled && !st->ctx->PointSizeIsSet) {
153       if (new_gp) {
154          st->dirty |= ST_NEW_GS_CONSTANTS;
155       } else if (new_tep) {
156          st->dirty |= ST_NEW_TES_CONSTANTS;
157       } else {
158          st->dirty |= ST_NEW_VS_CONSTANTS;
159       }
160    }
161    st->ctx->LastVertexStageDirty = false;
162 
163    st->dirty |= dirty;
164 }
165 
st_update_edgeflags(struct st_context * st,bool per_vertex_edgeflags)166 void st_update_edgeflags(struct st_context *st, bool per_vertex_edgeflags)
167 {
168    bool edgeflags_enabled = st->ctx->Polygon.FrontMode != GL_FILL ||
169                             st->ctx->Polygon.BackMode != GL_FILL;
170    bool vertdata_edgeflags = edgeflags_enabled && per_vertex_edgeflags;
171 
172    if (vertdata_edgeflags != st->vertdata_edgeflags) {
173       st->vertdata_edgeflags = vertdata_edgeflags;
174 
175       struct gl_program *vp = st->ctx->VertexProgram._Current;
176       if (vp)
177          st->dirty |= ST_NEW_VERTEX_PROGRAM(st, vp);
178    }
179 
180    bool edgeflag_culls_prims = edgeflags_enabled && !vertdata_edgeflags &&
181                                !st->ctx->Current.Attrib[VERT_ATTRIB_EDGEFLAG][0];
182    if (edgeflag_culls_prims != st->edgeflag_culls_prims) {
183       st->edgeflag_culls_prims = edgeflag_culls_prims;
184       st->dirty |= ST_NEW_RASTERIZER;
185    }
186 }
187 
check_attrib_edgeflag(struct st_context * st)188 static void check_attrib_edgeflag(struct st_context *st)
189 {
190    st_update_edgeflags(st, _mesa_draw_edge_flag_array_enabled(st->ctx));
191 }
192 
193 /***********************************************************************
194  * Update all derived state:
195  */
196 
st_validate_state(struct st_context * st,enum st_pipeline pipeline)197 void st_validate_state( struct st_context *st, enum st_pipeline pipeline )
198 {
199    struct gl_context *ctx = st->ctx;
200    uint64_t dirty, pipeline_mask;
201    uint32_t dirty_lo, dirty_hi;
202 
203    /* Get Mesa driver state.
204     *
205     * Inactive states are shader states not used by shaders at the moment.
206     */
207    st->dirty |= ctx->NewDriverState & st->active_states & ST_ALL_STATES_MASK;
208    ctx->NewDriverState &= ~st->dirty;
209 
210    /* Get pipeline state. */
211    switch (pipeline) {
212    case ST_PIPELINE_RENDER:
213    case ST_PIPELINE_RENDER_NO_VARRAYS:
214       if (st->ctx->API == API_OPENGL_COMPAT)
215          check_attrib_edgeflag(st);
216 
217       if (st->gfx_shaders_may_be_dirty) {
218          check_program_state(st);
219          st->gfx_shaders_may_be_dirty = false;
220       }
221 
222       st_manager_validate_framebuffers(st);
223 
224       if (pipeline == ST_PIPELINE_RENDER)
225          pipeline_mask = ST_PIPELINE_RENDER_STATE_MASK;
226       else
227          pipeline_mask = ST_PIPELINE_RENDER_STATE_MASK_NO_VARRAYS;
228       break;
229 
230    case ST_PIPELINE_CLEAR:
231       st_manager_validate_framebuffers(st);
232       pipeline_mask = ST_PIPELINE_CLEAR_STATE_MASK;
233       break;
234 
235    case ST_PIPELINE_META:
236       if (st->gfx_shaders_may_be_dirty) {
237          check_program_state(st);
238          st->gfx_shaders_may_be_dirty = false;
239       }
240 
241       st_manager_validate_framebuffers(st);
242       pipeline_mask = ST_PIPELINE_META_STATE_MASK;
243       break;
244 
245    case ST_PIPELINE_UPDATE_FRAMEBUFFER:
246       st_manager_validate_framebuffers(st);
247       pipeline_mask = ST_PIPELINE_UPDATE_FB_STATE_MASK;
248       break;
249 
250    case ST_PIPELINE_COMPUTE: {
251       struct gl_program *old_cp = st->cp;
252       struct gl_program *new_cp = ctx->ComputeProgram._Current;
253 
254       if (new_cp != old_cp) {
255          if (old_cp)
256             st->dirty |= old_cp->affected_states;
257          assert(new_cp);
258          st->dirty |= new_cp->affected_states;
259       }
260 
261       st->compute_shader_may_be_dirty = false;
262 
263       /*
264        * We add the ST_NEW_FB_STATE bit here as well, because glBindFramebuffer
265        * acts as a barrier that breaks feedback loops between the framebuffer
266        * and textures bound to the framebuffer, even when those textures are
267        * accessed by compute shaders; so we must inform the driver of new
268        * framebuffer state.
269        */
270       pipeline_mask = ST_PIPELINE_COMPUTE_STATE_MASK | ST_NEW_FB_STATE;
271       break;
272    }
273 
274    default:
275       unreachable("Invalid pipeline specified");
276    }
277 
278    dirty = st->dirty & pipeline_mask;
279    if (!dirty)
280       return;
281 
282    dirty_lo = dirty;
283    dirty_hi = dirty >> 32;
284 
285    /* Update states.
286     *
287     * Don't use u_bit_scan64, it may be slower on 32-bit.
288     */
289    while (dirty_lo)
290       update_functions[u_bit_scan(&dirty_lo)](st);
291    while (dirty_hi)
292       update_functions[32 + u_bit_scan(&dirty_hi)](st);
293 
294    /* Clear the render or compute state bits. */
295    st->dirty &= ~pipeline_mask;
296 }
297