• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /****************************************************************************
2  * Copyright (C) 2015 Intel Corporation.   All Rights Reserved.
3  *
4  * Permission is hereby granted, free of charge, to any person obtaining a
5  * copy of this software and associated documentation files (the "Software"),
6  * to deal in the Software without restriction, including without limitation
7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8  * and/or sell copies of the Software, and to permit persons to whom the
9  * Software is furnished to do so, subject to the following conditions:
10  *
11  * The above copyright notice and this permission notice (including the next
12  * paragraph) shall be included in all copies or substantial portions of the
13  * Software.
14  *
15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
21  * IN THE SOFTWARE.
22  ***************************************************************************/
23 
24 #include <llvm/Config/llvm-config.h>
25 
26 #if LLVM_VERSION_MAJOR < 7
27 // llvm redefines DEBUG
28 #pragma push_macro("DEBUG")
29 #undef DEBUG
30 #endif
31 
32 #include <rasterizer/core/state.h>
33 #include "JitManager.h"
34 
35 #if LLVM_VERSION_MAJOR < 7
36 #pragma pop_macro("DEBUG")
37 #endif
38 
39 #include "common/os.h"
40 #include "jit_api.h"
41 #include "gen_state_llvm.h"
42 #include "core/multisample.h"
43 #include "core/state_funcs.h"
44 
45 #include "gallivm/lp_bld_tgsi.h"
46 #include "util/format/u_format.h"
47 
48 #include "util/u_memory.h"
49 #include "util/u_inlines.h"
50 #include "util/u_helpers.h"
51 #include "util/u_framebuffer.h"
52 #include "util/u_viewport.h"
53 #include "util/u_prim.h"
54 
55 #include "swr_state.h"
56 #include "swr_context.h"
57 #include "gen_surf_state_llvm.h"
58 #include "gen_swr_context_llvm.h"
59 #include "swr_screen.h"
60 #include "swr_resource.h"
61 #include "swr_tex_sample.h"
62 #include "swr_scratch.h"
63 #include "swr_shader.h"
64 #include "swr_fence.h"
65 
66 /* These should be pulled out into separate files as necessary
67  * Just initializing everything here to get going. */
68 
69 static void *
swr_create_blend_state(struct pipe_context * pipe,const struct pipe_blend_state * blend)70 swr_create_blend_state(struct pipe_context *pipe,
71                        const struct pipe_blend_state *blend)
72 {
73    struct swr_blend_state *state = CALLOC_STRUCT(swr_blend_state);
74    assert(state != nullptr);
75 
76    memcpy(&state->pipe, blend, sizeof(*blend));
77 
78    struct pipe_blend_state *pipe_blend = &state->pipe;
79 
80    for (int target = 0;
81         target < std::min(SWR_NUM_RENDERTARGETS, PIPE_MAX_COLOR_BUFS);
82         target++) {
83 
84       struct pipe_rt_blend_state *rt_blend = &pipe_blend->rt[target];
85       SWR_RENDER_TARGET_BLEND_STATE &blendState =
86          state->blendState.renderTarget[target];
87       RENDER_TARGET_BLEND_COMPILE_STATE &compileState =
88          state->compileState[target];
89 
90       if (target != 0 && !pipe_blend->independent_blend_enable) {
91          memcpy(&compileState,
92                 &state->compileState[0],
93                 sizeof(RENDER_TARGET_BLEND_COMPILE_STATE));
94          continue;
95       }
96 
97       compileState.blendEnable = rt_blend->blend_enable;
98       if (compileState.blendEnable) {
99          compileState.sourceAlphaBlendFactor =
100             swr_convert_blend_factor(rt_blend->alpha_src_factor);
101          compileState.destAlphaBlendFactor =
102             swr_convert_blend_factor(rt_blend->alpha_dst_factor);
103          compileState.sourceBlendFactor =
104             swr_convert_blend_factor(rt_blend->rgb_src_factor);
105          compileState.destBlendFactor =
106             swr_convert_blend_factor(rt_blend->rgb_dst_factor);
107 
108          compileState.colorBlendFunc =
109             swr_convert_blend_func(rt_blend->rgb_func);
110          compileState.alphaBlendFunc =
111             swr_convert_blend_func(rt_blend->alpha_func);
112       }
113       compileState.logicOpEnable = state->pipe.logicop_enable;
114       if (compileState.logicOpEnable) {
115          compileState.logicOpFunc =
116             swr_convert_logic_op(state->pipe.logicop_func);
117       }
118 
119       blendState.writeDisableRed =
120          (rt_blend->colormask & PIPE_MASK_R) ? 0 : 1;
121       blendState.writeDisableGreen =
122          (rt_blend->colormask & PIPE_MASK_G) ? 0 : 1;
123       blendState.writeDisableBlue =
124          (rt_blend->colormask & PIPE_MASK_B) ? 0 : 1;
125       blendState.writeDisableAlpha =
126          (rt_blend->colormask & PIPE_MASK_A) ? 0 : 1;
127 
128       if (rt_blend->colormask == 0)
129          compileState.blendEnable = false;
130    }
131 
132    return state;
133 }
134 
135 static void
swr_bind_blend_state(struct pipe_context * pipe,void * blend)136 swr_bind_blend_state(struct pipe_context *pipe, void *blend)
137 {
138    struct swr_context *ctx = swr_context(pipe);
139 
140    if (ctx->blend == blend)
141       return;
142 
143    ctx->blend = (swr_blend_state *)blend;
144 
145    ctx->dirty |= SWR_NEW_BLEND;
146 }
147 
148 static void
swr_delete_blend_state(struct pipe_context * pipe,void * blend)149 swr_delete_blend_state(struct pipe_context *pipe, void *blend)
150 {
151    FREE(blend);
152 }
153 
154 static void
swr_set_blend_color(struct pipe_context * pipe,const struct pipe_blend_color * color)155 swr_set_blend_color(struct pipe_context *pipe,
156                     const struct pipe_blend_color *color)
157 {
158    struct swr_context *ctx = swr_context(pipe);
159 
160    ctx->blend_color = *color;
161 
162    ctx->dirty |= SWR_NEW_BLEND;
163 }
164 
165 static void
swr_set_stencil_ref(struct pipe_context * pipe,const struct pipe_stencil_ref * ref)166 swr_set_stencil_ref(struct pipe_context *pipe,
167                     const struct pipe_stencil_ref *ref)
168 {
169    struct swr_context *ctx = swr_context(pipe);
170 
171    ctx->stencil_ref = *ref;
172 
173    ctx->dirty |= SWR_NEW_DEPTH_STENCIL_ALPHA;
174 }
175 
176 static void *
swr_create_depth_stencil_state(struct pipe_context * pipe,const struct pipe_depth_stencil_alpha_state * depth_stencil)177 swr_create_depth_stencil_state(
178    struct pipe_context *pipe,
179    const struct pipe_depth_stencil_alpha_state *depth_stencil)
180 {
181    struct pipe_depth_stencil_alpha_state *state;
182 
183    state = (pipe_depth_stencil_alpha_state *)mem_dup(depth_stencil,
184                                                      sizeof *depth_stencil);
185 
186    return state;
187 }
188 
189 static void
swr_bind_depth_stencil_state(struct pipe_context * pipe,void * depth_stencil)190 swr_bind_depth_stencil_state(struct pipe_context *pipe, void *depth_stencil)
191 {
192    struct swr_context *ctx = swr_context(pipe);
193 
194    if (ctx->depth_stencil == (pipe_depth_stencil_alpha_state *)depth_stencil)
195       return;
196 
197    ctx->depth_stencil = (pipe_depth_stencil_alpha_state *)depth_stencil;
198 
199    ctx->dirty |= SWR_NEW_DEPTH_STENCIL_ALPHA;
200 }
201 
202 static void
swr_delete_depth_stencil_state(struct pipe_context * pipe,void * depth)203 swr_delete_depth_stencil_state(struct pipe_context *pipe, void *depth)
204 {
205    FREE(depth);
206 }
207 
208 
209 static void *
swr_create_rasterizer_state(struct pipe_context * pipe,const struct pipe_rasterizer_state * rast)210 swr_create_rasterizer_state(struct pipe_context *pipe,
211                             const struct pipe_rasterizer_state *rast)
212 {
213    struct pipe_rasterizer_state *state;
214    state = (pipe_rasterizer_state *)mem_dup(rast, sizeof *rast);
215 
216    return state;
217 }
218 
219 static void
swr_bind_rasterizer_state(struct pipe_context * pipe,void * handle)220 swr_bind_rasterizer_state(struct pipe_context *pipe, void *handle)
221 {
222    struct swr_context *ctx = swr_context(pipe);
223    const struct pipe_rasterizer_state *rasterizer =
224       (const struct pipe_rasterizer_state *)handle;
225 
226    if (ctx->rasterizer == (pipe_rasterizer_state *)rasterizer)
227       return;
228 
229    ctx->rasterizer = (pipe_rasterizer_state *)rasterizer;
230 
231    ctx->dirty |= SWR_NEW_RASTERIZER;
232 }
233 
234 static void
swr_delete_rasterizer_state(struct pipe_context * pipe,void * rasterizer)235 swr_delete_rasterizer_state(struct pipe_context *pipe, void *rasterizer)
236 {
237    FREE(rasterizer);
238 }
239 
240 
241 static void *
swr_create_sampler_state(struct pipe_context * pipe,const struct pipe_sampler_state * sampler)242 swr_create_sampler_state(struct pipe_context *pipe,
243                          const struct pipe_sampler_state *sampler)
244 {
245    struct pipe_sampler_state *state =
246       (pipe_sampler_state *)mem_dup(sampler, sizeof *sampler);
247 
248    return state;
249 }
250 
251 static void
swr_bind_sampler_states(struct pipe_context * pipe,enum pipe_shader_type shader,unsigned start,unsigned num,void ** samplers)252 swr_bind_sampler_states(struct pipe_context *pipe,
253                         enum pipe_shader_type shader,
254                         unsigned start,
255                         unsigned num,
256                         void **samplers)
257 {
258    struct swr_context *ctx = swr_context(pipe);
259    unsigned i;
260 
261    assert(shader < PIPE_SHADER_TYPES);
262    assert(start + num <= ARRAY_SIZE(ctx->samplers[shader]));
263 
264    /* set the new samplers */
265    ctx->num_samplers[shader] = num;
266    for (i = 0; i < num; i++) {
267       ctx->samplers[shader][start + i] = (pipe_sampler_state *)samplers[i];
268    }
269 
270    ctx->dirty |= SWR_NEW_SAMPLER;
271 }
272 
273 static void
swr_delete_sampler_state(struct pipe_context * pipe,void * sampler)274 swr_delete_sampler_state(struct pipe_context *pipe, void *sampler)
275 {
276    FREE(sampler);
277 }
278 
279 
280 static struct pipe_sampler_view *
swr_create_sampler_view(struct pipe_context * pipe,struct pipe_resource * texture,const struct pipe_sampler_view * templ)281 swr_create_sampler_view(struct pipe_context *pipe,
282                         struct pipe_resource *texture,
283                         const struct pipe_sampler_view *templ)
284 {
285    struct pipe_sampler_view *view = CALLOC_STRUCT(pipe_sampler_view);
286 
287    if (view) {
288       *view = *templ;
289       view->reference.count = 1;
290       view->texture = NULL;
291       pipe_resource_reference(&view->texture, texture);
292       view->context = pipe;
293    }
294 
295    return view;
296 }
297 
298 static void
swr_set_sampler_views(struct pipe_context * pipe,enum pipe_shader_type shader,unsigned start,unsigned num,struct pipe_sampler_view ** views)299 swr_set_sampler_views(struct pipe_context *pipe,
300                       enum pipe_shader_type shader,
301                       unsigned start,
302                       unsigned num,
303                       struct pipe_sampler_view **views)
304 {
305    struct swr_context *ctx = swr_context(pipe);
306    uint i;
307 
308    assert(num <= PIPE_MAX_SHADER_SAMPLER_VIEWS);
309 
310    assert(shader < PIPE_SHADER_TYPES);
311    assert(start + num <= ARRAY_SIZE(ctx->sampler_views[shader]));
312 
313    /* set the new sampler views */
314    ctx->num_sampler_views[shader] = num;
315    for (i = 0; i < num; i++) {
316       pipe_sampler_view_reference(&ctx->sampler_views[shader][start + i],
317                                   views[i]);
318    }
319 
320    ctx->dirty |= SWR_NEW_SAMPLER_VIEW;
321 }
322 
323 static void
swr_sampler_view_destroy(struct pipe_context * pipe,struct pipe_sampler_view * view)324 swr_sampler_view_destroy(struct pipe_context *pipe,
325                          struct pipe_sampler_view *view)
326 {
327    pipe_resource_reference(&view->texture, NULL);
328    FREE(view);
329 }
330 
331 static void *
swr_create_vs_state(struct pipe_context * pipe,const struct pipe_shader_state * vs)332 swr_create_vs_state(struct pipe_context *pipe,
333                     const struct pipe_shader_state *vs)
334 {
335    struct swr_vertex_shader *swr_vs = new swr_vertex_shader;
336    if (!swr_vs)
337       return NULL;
338 
339    swr_vs->pipe.tokens = tgsi_dup_tokens(vs->tokens);
340    swr_vs->pipe.stream_output = vs->stream_output;
341 
342    lp_build_tgsi_info(vs->tokens, &swr_vs->info);
343 
344    swr_vs->soState = {0};
345 
346    if (swr_vs->pipe.stream_output.num_outputs) {
347       pipe_stream_output_info *stream_output = &swr_vs->pipe.stream_output;
348 
349       swr_vs->soState.soEnable = true;
350       // soState.rasterizerDisable set on state dirty
351       // soState.streamToRasterizer not used
352 
353       for (uint32_t i = 0; i < stream_output->num_outputs; i++) {
354          unsigned attrib_slot = stream_output->output[i].register_index;
355          attrib_slot = swr_so_adjust_attrib(attrib_slot, swr_vs);
356          swr_vs->soState.streamMasks[stream_output->output[i].stream] |=
357             (1 << attrib_slot);
358       }
359       for (uint32_t i = 0; i < MAX_SO_STREAMS; i++) {
360         swr_vs->soState.streamNumEntries[i] =
361              _mm_popcnt_u32(swr_vs->soState.streamMasks[i]);
362        }
363    }
364 
365    return swr_vs;
366 }
367 
368 static void
swr_bind_vs_state(struct pipe_context * pipe,void * vs)369 swr_bind_vs_state(struct pipe_context *pipe, void *vs)
370 {
371    struct swr_context *ctx = swr_context(pipe);
372 
373    if (ctx->vs == vs)
374       return;
375 
376    ctx->vs = (swr_vertex_shader *)vs;
377    ctx->dirty |= SWR_NEW_VS;
378 }
379 
380 static void
swr_delete_vs_state(struct pipe_context * pipe,void * vs)381 swr_delete_vs_state(struct pipe_context *pipe, void *vs)
382 {
383    struct swr_vertex_shader *swr_vs = (swr_vertex_shader *)vs;
384    FREE((void *)swr_vs->pipe.tokens);
385    struct swr_screen *screen = swr_screen(pipe->screen);
386 
387    /* Defer deletion of vs state */
388    swr_fence_work_delete_vs(screen->flush_fence, swr_vs);
389 }
390 
391 static void *
swr_create_fs_state(struct pipe_context * pipe,const struct pipe_shader_state * fs)392 swr_create_fs_state(struct pipe_context *pipe,
393                     const struct pipe_shader_state *fs)
394 {
395    struct swr_fragment_shader *swr_fs = new swr_fragment_shader;
396    if (!swr_fs)
397       return NULL;
398 
399    swr_fs->pipe.tokens = tgsi_dup_tokens(fs->tokens);
400 
401    lp_build_tgsi_info(fs->tokens, &swr_fs->info);
402 
403    return swr_fs;
404 }
405 
406 
407 static void
swr_bind_fs_state(struct pipe_context * pipe,void * fs)408 swr_bind_fs_state(struct pipe_context *pipe, void *fs)
409 {
410    struct swr_context *ctx = swr_context(pipe);
411 
412    if (ctx->fs == fs)
413       return;
414 
415    ctx->fs = (swr_fragment_shader *)fs;
416    ctx->dirty |= SWR_NEW_FS;
417 }
418 
419 static void
swr_delete_fs_state(struct pipe_context * pipe,void * fs)420 swr_delete_fs_state(struct pipe_context *pipe, void *fs)
421 {
422    struct swr_fragment_shader *swr_fs = (swr_fragment_shader *)fs;
423    FREE((void *)swr_fs->pipe.tokens);
424    struct swr_screen *screen = swr_screen(pipe->screen);
425 
426    /* Defer deleton of fs state */
427    swr_fence_work_delete_fs(screen->flush_fence, swr_fs);
428 }
429 
430 static void *
swr_create_gs_state(struct pipe_context * pipe,const struct pipe_shader_state * gs)431 swr_create_gs_state(struct pipe_context *pipe,
432                     const struct pipe_shader_state *gs)
433 {
434    struct swr_geometry_shader *swr_gs = new swr_geometry_shader;
435    if (!swr_gs)
436       return NULL;
437 
438    swr_gs->pipe.tokens = tgsi_dup_tokens(gs->tokens);
439    lp_build_tgsi_info(gs->tokens, &swr_gs->info);
440    return swr_gs;
441 }
442 
443 static void
swr_bind_gs_state(struct pipe_context * pipe,void * gs)444 swr_bind_gs_state(struct pipe_context *pipe, void *gs)
445 {
446    struct swr_context *ctx = swr_context(pipe);
447 
448    if (ctx->gs == gs)
449       return;
450 
451    ctx->gs = (swr_geometry_shader *)gs;
452    ctx->dirty |= SWR_NEW_GS;
453 }
454 
455 static void
swr_delete_gs_state(struct pipe_context * pipe,void * gs)456 swr_delete_gs_state(struct pipe_context *pipe, void *gs)
457 {
458    struct swr_geometry_shader *swr_gs = (swr_geometry_shader *)gs;
459    FREE((void *)swr_gs->pipe.tokens);
460    struct swr_screen *screen = swr_screen(pipe->screen);
461 
462    /* Defer deleton of fs state */
463    swr_fence_work_delete_gs(screen->flush_fence, swr_gs);
464 }
465 
466 static void *
swr_create_tcs_state(struct pipe_context * pipe,const struct pipe_shader_state * tcs)467 swr_create_tcs_state(struct pipe_context *pipe,
468                      const struct pipe_shader_state *tcs)
469 {
470    struct swr_tess_control_shader *swr_tcs = new swr_tess_control_shader;
471    if (!swr_tcs)
472       return NULL;
473 
474    swr_tcs->pipe.tokens = tgsi_dup_tokens(tcs->tokens);
475    lp_build_tgsi_info(tcs->tokens, &swr_tcs->info);
476    return swr_tcs;
477 }
478 
479 static void
swr_bind_tcs_state(struct pipe_context * pipe,void * tcs)480 swr_bind_tcs_state(struct pipe_context *pipe, void *tcs)
481 {
482    struct swr_context *ctx = swr_context(pipe);
483 
484    if (ctx->tcs == tcs)
485       return;
486 
487    ctx->tcs = (swr_tess_control_shader *)tcs;
488    ctx->dirty |= SWR_NEW_TCS;
489    ctx->dirty |= SWR_NEW_TS;
490 }
491 
492 static void
swr_delete_tcs_state(struct pipe_context * pipe,void * tcs)493 swr_delete_tcs_state(struct pipe_context *pipe, void *tcs)
494 {
495    struct swr_tess_control_shader *swr_tcs = (swr_tess_control_shader *)tcs;
496    FREE((void *)swr_tcs->pipe.tokens);
497    struct swr_screen *screen = swr_screen(pipe->screen);
498 
499    /* Defer deleton of tcs state */
500    swr_fence_work_delete_tcs(screen->flush_fence, swr_tcs);
501 }
502 
503 static void *
swr_create_tes_state(struct pipe_context * pipe,const struct pipe_shader_state * tes)504 swr_create_tes_state(struct pipe_context *pipe,
505                      const struct pipe_shader_state *tes)
506 {
507    struct swr_tess_evaluation_shader *swr_tes = new swr_tess_evaluation_shader;
508    if (!swr_tes)
509       return NULL;
510 
511    swr_tes->pipe.tokens = tgsi_dup_tokens(tes->tokens);
512    lp_build_tgsi_info(tes->tokens, &swr_tes->info);
513    return swr_tes;
514 }
515 
516 static void
swr_bind_tes_state(struct pipe_context * pipe,void * tes)517 swr_bind_tes_state(struct pipe_context *pipe, void *tes)
518 {
519    struct swr_context *ctx = swr_context(pipe);
520 
521    if (ctx->tes == tes)
522       return;
523 
524    // Save current tessellator state first
525    if (ctx->tes != nullptr) {
526       ctx->tes->ts_state = ctx->tsState;
527    }
528 
529    ctx->tes = (swr_tess_evaluation_shader *)tes;
530 
531    ctx->dirty |= SWR_NEW_TES;
532    ctx->dirty |= SWR_NEW_TS;
533 }
534 
535 static void
swr_delete_tes_state(struct pipe_context * pipe,void * tes)536 swr_delete_tes_state(struct pipe_context *pipe, void *tes)
537 {
538    struct swr_tess_evaluation_shader *swr_tes = (swr_tess_evaluation_shader *)tes;
539    FREE((void *)swr_tes->pipe.tokens);
540    struct swr_screen *screen = swr_screen(pipe->screen);
541 
542    /* Defer deleton of tes state */
543    swr_fence_work_delete_tes(screen->flush_fence, swr_tes);
544 }
545 
546 static void
swr_set_constant_buffer(struct pipe_context * pipe,enum pipe_shader_type shader,uint index,const struct pipe_constant_buffer * cb)547 swr_set_constant_buffer(struct pipe_context *pipe,
548                         enum pipe_shader_type shader,
549                         uint index,
550                         const struct pipe_constant_buffer *cb)
551 {
552    struct swr_context *ctx = swr_context(pipe);
553    struct pipe_resource *constants = cb ? cb->buffer : NULL;
554 
555    assert(shader < PIPE_SHADER_TYPES);
556    assert(index < ARRAY_SIZE(ctx->constants[shader]));
557 
558    /* note: reference counting */
559    util_copy_constant_buffer(&ctx->constants[shader][index], cb);
560 
561    if (shader == PIPE_SHADER_VERTEX) {
562       ctx->dirty |= SWR_NEW_VSCONSTANTS;
563    } else if (shader == PIPE_SHADER_FRAGMENT) {
564       ctx->dirty |= SWR_NEW_FSCONSTANTS;
565    } else if (shader == PIPE_SHADER_GEOMETRY) {
566       ctx->dirty |= SWR_NEW_GSCONSTANTS;
567    } else if (shader == PIPE_SHADER_TESS_CTRL) {
568       ctx->dirty |= SWR_NEW_TCSCONSTANTS;
569    } else if (shader == PIPE_SHADER_TESS_EVAL) {
570       ctx->dirty |= SWR_NEW_TESCONSTANTS;
571    }
572    if (cb && cb->user_buffer) {
573       pipe_resource_reference(&constants, NULL);
574    }
575 }
576 
577 
578 static void *
swr_create_vertex_elements_state(struct pipe_context * pipe,unsigned num_elements,const struct pipe_vertex_element * attribs)579 swr_create_vertex_elements_state(struct pipe_context *pipe,
580                                  unsigned num_elements,
581                                  const struct pipe_vertex_element *attribs)
582 {
583    struct swr_vertex_element_state *velems;
584    assert(num_elements <= PIPE_MAX_ATTRIBS);
585    velems = new swr_vertex_element_state;
586    if (velems) {
587       memset((void*)&velems->fsState, 0, sizeof(velems->fsState));
588       velems->fsState.bVertexIDOffsetEnable = true;
589       velems->fsState.numAttribs = num_elements;
590       for (unsigned i = 0; i < num_elements; i++) {
591          // XXX: we should do this keyed on the VS usage info
592 
593          const struct util_format_description *desc =
594             util_format_description(attribs[i].src_format);
595 
596          velems->fsState.layout[i].AlignedByteOffset = attribs[i].src_offset;
597          velems->fsState.layout[i].Format =
598             mesa_to_swr_format(attribs[i].src_format);
599          velems->fsState.layout[i].StreamIndex =
600             attribs[i].vertex_buffer_index;
601          velems->fsState.layout[i].InstanceEnable =
602             attribs[i].instance_divisor != 0;
603          velems->fsState.layout[i].ComponentControl0 =
604             desc->channel[0].type != UTIL_FORMAT_TYPE_VOID
605             ? ComponentControl::StoreSrc
606             : ComponentControl::Store0;
607          velems->fsState.layout[i].ComponentControl1 =
608             desc->channel[1].type != UTIL_FORMAT_TYPE_VOID
609             ? ComponentControl::StoreSrc
610             : ComponentControl::Store0;
611          velems->fsState.layout[i].ComponentControl2 =
612             desc->channel[2].type != UTIL_FORMAT_TYPE_VOID
613             ? ComponentControl::StoreSrc
614             : ComponentControl::Store0;
615          velems->fsState.layout[i].ComponentControl3 =
616             desc->channel[3].type != UTIL_FORMAT_TYPE_VOID
617             ? ComponentControl::StoreSrc
618             : ComponentControl::Store1Fp;
619          velems->fsState.layout[i].ComponentPacking = ComponentEnable::XYZW;
620          velems->fsState.layout[i].InstanceAdvancementState =
621             attribs[i].instance_divisor;
622 
623          /* Calculate the pitch of each stream */
624          const SWR_FORMAT_INFO &swr_desc = GetFormatInfo(
625             mesa_to_swr_format(attribs[i].src_format));
626          velems->stream_pitch[attribs[i].vertex_buffer_index] += swr_desc.Bpp;
627 
628          if (attribs[i].instance_divisor != 0) {
629             velems->instanced_bufs |= 1U << attribs[i].vertex_buffer_index;
630             uint32_t *min_instance_div =
631                &velems->min_instance_div[attribs[i].vertex_buffer_index];
632             if (!*min_instance_div ||
633                 attribs[i].instance_divisor < *min_instance_div)
634                *min_instance_div = attribs[i].instance_divisor;
635          }
636       }
637    }
638 
639    return velems;
640 }
641 
642 static void
swr_bind_vertex_elements_state(struct pipe_context * pipe,void * velems)643 swr_bind_vertex_elements_state(struct pipe_context *pipe, void *velems)
644 {
645    struct swr_context *ctx = swr_context(pipe);
646    struct swr_vertex_element_state *swr_velems =
647       (struct swr_vertex_element_state *)velems;
648 
649    ctx->velems = swr_velems;
650    ctx->dirty |= SWR_NEW_VERTEX;
651 }
652 
653 static void
swr_delete_vertex_elements_state(struct pipe_context * pipe,void * velems)654 swr_delete_vertex_elements_state(struct pipe_context *pipe, void *velems)
655 {
656    struct swr_vertex_element_state *swr_velems =
657       (struct swr_vertex_element_state *) velems;
658    /* XXX Need to destroy fetch shader? */
659    delete swr_velems;
660 }
661 
662 
663 static void
swr_set_vertex_buffers(struct pipe_context * pipe,unsigned start_slot,unsigned num_elements,const struct pipe_vertex_buffer * buffers)664 swr_set_vertex_buffers(struct pipe_context *pipe,
665                        unsigned start_slot,
666                        unsigned num_elements,
667                        const struct pipe_vertex_buffer *buffers)
668 {
669    struct swr_context *ctx = swr_context(pipe);
670 
671    assert(num_elements <= PIPE_MAX_ATTRIBS);
672 
673    util_set_vertex_buffers_count(ctx->vertex_buffer,
674                                  &ctx->num_vertex_buffers,
675                                  buffers,
676                                  start_slot,
677                                  num_elements);
678 
679    ctx->dirty |= SWR_NEW_VERTEX;
680 }
681 
682 
683 static void
swr_set_polygon_stipple(struct pipe_context * pipe,const struct pipe_poly_stipple * stipple)684 swr_set_polygon_stipple(struct pipe_context *pipe,
685                         const struct pipe_poly_stipple *stipple)
686 {
687    struct swr_context *ctx = swr_context(pipe);
688 
689    ctx->poly_stipple.pipe = *stipple; /* struct copy */
690    ctx->dirty |= SWR_NEW_STIPPLE;
691 }
692 
693 static void
swr_set_clip_state(struct pipe_context * pipe,const struct pipe_clip_state * clip)694 swr_set_clip_state(struct pipe_context *pipe,
695                    const struct pipe_clip_state *clip)
696 {
697    struct swr_context *ctx = swr_context(pipe);
698 
699    ctx->clip = *clip;
700    /* XXX Unimplemented, but prevents crash */
701 
702    ctx->dirty |= SWR_NEW_CLIP;
703 }
704 
705 
706 static void
swr_set_scissor_states(struct pipe_context * pipe,unsigned start_slot,unsigned num_scissors,const struct pipe_scissor_state * scissors)707 swr_set_scissor_states(struct pipe_context *pipe,
708                        unsigned start_slot,
709                        unsigned num_scissors,
710                        const struct pipe_scissor_state *scissors)
711 {
712    struct swr_context *ctx = swr_context(pipe);
713 
714    memcpy(ctx->scissors + start_slot, scissors,
715           sizeof(struct pipe_scissor_state) * num_scissors);
716 
717    for (unsigned i = 0; i < num_scissors; i++) {
718       auto idx = start_slot + i;
719       ctx->swr_scissors[idx].xmin = scissors[idx].minx;
720       ctx->swr_scissors[idx].xmax = scissors[idx].maxx;
721       ctx->swr_scissors[idx].ymin = scissors[idx].miny;
722       ctx->swr_scissors[idx].ymax = scissors[idx].maxy;
723    }
724    ctx->dirty |= SWR_NEW_SCISSOR;
725 }
726 
727 static void
swr_set_viewport_states(struct pipe_context * pipe,unsigned start_slot,unsigned num_viewports,const struct pipe_viewport_state * vpt)728 swr_set_viewport_states(struct pipe_context *pipe,
729                         unsigned start_slot,
730                         unsigned num_viewports,
731                         const struct pipe_viewport_state *vpt)
732 {
733    struct swr_context *ctx = swr_context(pipe);
734 
735    memcpy(ctx->viewports + start_slot, vpt, sizeof(struct pipe_viewport_state) * num_viewports);
736    ctx->dirty |= SWR_NEW_VIEWPORT;
737 }
738 
739 
740 static void
swr_set_framebuffer_state(struct pipe_context * pipe,const struct pipe_framebuffer_state * fb)741 swr_set_framebuffer_state(struct pipe_context *pipe,
742                           const struct pipe_framebuffer_state *fb)
743 {
744    struct swr_context *ctx = swr_context(pipe);
745 
746    bool changed = !util_framebuffer_state_equal(&ctx->framebuffer, fb);
747 
748    assert(fb->width <= KNOB_GUARDBAND_WIDTH);
749    assert(fb->height <= KNOB_GUARDBAND_HEIGHT);
750 
751    if (changed) {
752       util_copy_framebuffer_state(&ctx->framebuffer, fb);
753 
754       /* 0 and 1 both indicate no msaa.  Core doesn't understand 0 samples */
755       ctx->framebuffer.samples = std::max((ubyte)1, ctx->framebuffer.samples);
756 
757       ctx->dirty |= SWR_NEW_FRAMEBUFFER;
758    }
759 }
760 
761 
762 static void
swr_set_sample_mask(struct pipe_context * pipe,unsigned sample_mask)763 swr_set_sample_mask(struct pipe_context *pipe, unsigned sample_mask)
764 {
765    struct swr_context *ctx = swr_context(pipe);
766 
767    if (sample_mask != ctx->sample_mask) {
768       ctx->sample_mask = sample_mask;
769       ctx->dirty |= SWR_NEW_RASTERIZER;
770    }
771 }
772 
773 /*
774  * MSAA fixed sample position table
775  * used by update_derived and get_sample_position
776  * (integer locations on a 16x16 grid)
777  */
778 static const uint8_t swr_sample_positions[][2] =
779 { /* 1x*/ { 8, 8},
780   /* 2x*/ {12,12},{ 4, 4},
781   /* 4x*/ { 6, 2},{14, 6},{ 2,10},{10,14},
782   /* 8x*/ { 9, 5},{ 7,11},{13, 9},{ 5, 3},
783           { 3,13},{ 1, 7},{11,15},{15, 1},
784   /*16x*/ { 9, 9},{ 7, 5},{ 5,10},{12, 7},
785           { 3, 6},{10,13},{13,11},{11, 3},
786           { 6,14},{ 8, 1},{ 4, 2},{ 2,12},
787           { 0, 8},{15, 4},{14,15},{ 1, 0} };
788 
789 static void
swr_get_sample_position(struct pipe_context * pipe,unsigned sample_count,unsigned sample_index,float * out_value)790 swr_get_sample_position(struct pipe_context *pipe,
791                         unsigned sample_count, unsigned sample_index,
792                         float *out_value)
793 {
794    /* validate sample_count */
795    sample_count = GetNumSamples(GetSampleCount(sample_count));
796 
797    const uint8_t *sample = swr_sample_positions[sample_count-1 + sample_index];
798    out_value[0] = sample[0] / 16.0f;
799    out_value[1] = sample[1] / 16.0f;
800 }
801 
802 
803 /*
804  * Update resource in-use status
805  * All resources bound to color or depth targets marked as WRITE resources.
806  * VBO Vertex/index buffers and texture views marked as READ resources.
807  */
808 void
swr_update_resource_status(struct pipe_context * pipe,const struct pipe_draw_info * p_draw_info)809 swr_update_resource_status(struct pipe_context *pipe,
810                            const struct pipe_draw_info *p_draw_info)
811 {
812    struct swr_context *ctx = swr_context(pipe);
813    struct pipe_framebuffer_state *fb = &ctx->framebuffer;
814 
815    /* colorbuffer targets */
816    if (fb->nr_cbufs)
817       for (uint32_t i = 0; i < fb->nr_cbufs; ++i)
818          if (fb->cbufs[i])
819             swr_resource_write(fb->cbufs[i]->texture);
820 
821    /* depth/stencil target */
822    if (fb->zsbuf)
823       swr_resource_write(fb->zsbuf->texture);
824 
825    /* VBO vertex buffers */
826    for (uint32_t i = 0; i < ctx->num_vertex_buffers; i++) {
827       struct pipe_vertex_buffer *vb = &ctx->vertex_buffer[i];
828       if (!vb->is_user_buffer && vb->buffer.resource)
829          swr_resource_read(vb->buffer.resource);
830    }
831 
832    /* VBO index buffer */
833    if (p_draw_info && p_draw_info->index_size) {
834       if (!p_draw_info->has_user_indices)
835          swr_resource_read(p_draw_info->index.resource);
836    }
837 
838    /* transform feedback buffers */
839    for (uint32_t i = 0; i < ctx->num_so_targets; i++) {
840       struct pipe_stream_output_target *target = ctx->so_targets[i];
841       if (target && target->buffer)
842          swr_resource_write(target->buffer);
843    }
844 
845    /* texture sampler views */
846    for (uint32_t j : {PIPE_SHADER_VERTEX, PIPE_SHADER_FRAGMENT}) {
847       for (uint32_t i = 0; i < ctx->num_sampler_views[j]; i++) {
848          struct pipe_sampler_view *view = ctx->sampler_views[j][i];
849          if (view)
850             swr_resource_read(view->texture);
851       }
852    }
853 
854    /* constant buffers */
855    for (uint32_t j : {PIPE_SHADER_VERTEX, PIPE_SHADER_FRAGMENT}) {
856       for (uint32_t i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
857          struct pipe_constant_buffer *cb = &ctx->constants[j][i];
858          if (cb->buffer)
859             swr_resource_read(cb->buffer);
860       }
861    }
862 }
863 
864 static void
swr_update_texture_state(struct swr_context * ctx,enum pipe_shader_type shader_type,unsigned num_sampler_views,swr_jit_texture * textures)865 swr_update_texture_state(struct swr_context *ctx,
866                          enum pipe_shader_type shader_type,
867                          unsigned num_sampler_views,
868                          swr_jit_texture *textures)
869 {
870    for (unsigned i = 0; i < num_sampler_views; i++) {
871       struct pipe_sampler_view *view =
872          ctx->sampler_views[shader_type][i];
873       struct swr_jit_texture *jit_tex = &textures[i];
874 
875       memset(jit_tex, 0, sizeof(*jit_tex));
876       if (view) {
877          struct pipe_resource *res = view->texture;
878          struct swr_resource *swr_res = swr_resource(res);
879          SWR_SURFACE_STATE *swr = &swr_res->swr;
880          size_t *mip_offsets = swr_res->mip_offsets;
881          if (swr_res->has_depth && swr_res->has_stencil &&
882             !util_format_has_depth(util_format_description(view->format))) {
883             swr = &swr_res->secondary;
884             mip_offsets = swr_res->secondary_mip_offsets;
885          }
886 
887          jit_tex->width = res->width0;
888          jit_tex->height = res->height0;
889          jit_tex->base_ptr = (uint8_t*)swr->xpBaseAddress;
890          jit_tex->num_samples = swr->numSamples;
891          jit_tex->sample_stride = 0;
892          if (view->target != PIPE_BUFFER) {
893             jit_tex->first_level = view->u.tex.first_level;
894             jit_tex->last_level = view->u.tex.last_level;
895             if (view->target == PIPE_TEXTURE_3D)
896                jit_tex->depth = res->depth0;
897             else
898                jit_tex->depth =
899                   view->u.tex.last_layer - view->u.tex.first_layer + 1;
900             jit_tex->base_ptr += view->u.tex.first_layer *
901                swr->qpitch * swr->pitch;
902          } else {
903             unsigned view_blocksize = util_format_get_blocksize(view->format);
904             jit_tex->base_ptr += view->u.buf.offset;
905             jit_tex->width = view->u.buf.size / view_blocksize;
906             jit_tex->depth = 1;
907          }
908 
909          for (unsigned level = jit_tex->first_level;
910               level <= jit_tex->last_level;
911               level++) {
912             jit_tex->row_stride[level] = swr->pitch;
913             jit_tex->img_stride[level] = swr->qpitch * swr->pitch;
914             jit_tex->mip_offsets[level] = mip_offsets[level];
915          }
916       }
917    }
918 }
919 
920 static void
swr_update_sampler_state(struct swr_context * ctx,enum pipe_shader_type shader_type,unsigned num_samplers,swr_jit_sampler * samplers)921 swr_update_sampler_state(struct swr_context *ctx,
922                          enum pipe_shader_type shader_type,
923                          unsigned num_samplers,
924                          swr_jit_sampler *samplers)
925 {
926    for (unsigned i = 0; i < num_samplers; i++) {
927       const struct pipe_sampler_state *sampler =
928          ctx->samplers[shader_type][i];
929 
930       if (sampler) {
931          samplers[i].min_lod = sampler->min_lod;
932          samplers[i].max_lod = sampler->max_lod;
933          samplers[i].lod_bias = sampler->lod_bias;
934          COPY_4V(samplers[i].border_color, sampler->border_color.f);
935       }
936    }
937 }
938 
939 static void
swr_update_constants(struct swr_context * ctx,enum pipe_shader_type shaderType)940 swr_update_constants(struct swr_context *ctx, enum pipe_shader_type shaderType)
941 {
942    swr_draw_context *pDC = &ctx->swrDC;
943 
944    const float **constant;
945    uint32_t *num_constants;
946    struct swr_scratch_space *scratch;
947 
948    switch (shaderType) {
949    case PIPE_SHADER_VERTEX:
950       constant = pDC->constantVS;
951       num_constants = pDC->num_constantsVS;
952       scratch = &ctx->scratch->vs_constants;
953       break;
954    case PIPE_SHADER_FRAGMENT:
955       constant = pDC->constantFS;
956       num_constants = pDC->num_constantsFS;
957       scratch = &ctx->scratch->fs_constants;
958       break;
959    case PIPE_SHADER_GEOMETRY:
960       constant = pDC->constantGS;
961       num_constants = pDC->num_constantsGS;
962       scratch = &ctx->scratch->gs_constants;
963       break;
964    case PIPE_SHADER_TESS_CTRL:
965       constant = pDC->constantTCS;
966       num_constants = pDC->num_constantsTCS;
967       scratch = &ctx->scratch->tcs_constants;
968       break;
969    case PIPE_SHADER_TESS_EVAL:
970       constant = pDC->constantTES;
971       num_constants = pDC->num_constantsTES;
972       scratch = &ctx->scratch->tes_constants;
973       break;
974    default:
975       assert(0 && "Unsupported shader type constants");
976       return;
977    }
978 
979    for (UINT i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
980       const pipe_constant_buffer *cb = &ctx->constants[shaderType][i];
981       num_constants[i] = cb->buffer_size;
982       if (cb->buffer) {
983          constant[i] =
984             (const float *)(swr_resource_data(cb->buffer) +
985                             cb->buffer_offset);
986       } else {
987          /* Need to copy these constants to scratch space */
988          if (cb->user_buffer && cb->buffer_size) {
989             const void *ptr =
990                ((const uint8_t *)cb->user_buffer + cb->buffer_offset);
991             uint32_t size = AlignUp(cb->buffer_size, 4);
992             ptr = swr_copy_to_scratch_space(ctx, scratch, ptr, size);
993             constant[i] = (const float *)ptr;
994          }
995       }
996    }
997 }
998 
999 static bool
swr_change_rt(struct swr_context * ctx,unsigned attachment,const struct pipe_surface * sf)1000 swr_change_rt(struct swr_context *ctx,
1001               unsigned attachment,
1002               const struct pipe_surface *sf)
1003 {
1004    swr_draw_context *pDC = &ctx->swrDC;
1005    struct SWR_SURFACE_STATE *rt = &pDC->renderTargets[attachment];
1006 
1007    /* Do nothing if the render target hasn't changed */
1008    if ((!sf || !sf->texture) && (void*)(rt->xpBaseAddress) == nullptr)
1009       return false;
1010 
1011    /* Deal with disabling RT up front */
1012    if (!sf || !sf->texture) {
1013       /* If detaching attachment, mark tiles as RESOLVED so core
1014        * won't try to load from non-existent target. */
1015       swr_store_render_target(&ctx->pipe, attachment, SWR_TILE_RESOLVED);
1016       *rt = {0};
1017       return true;
1018    }
1019 
1020    const struct swr_resource *swr = swr_resource(sf->texture);
1021    const SWR_SURFACE_STATE *swr_surface = &swr->swr;
1022    SWR_FORMAT fmt = mesa_to_swr_format(sf->format);
1023 
1024    if (attachment == SWR_ATTACHMENT_STENCIL && swr->secondary.xpBaseAddress) {
1025       swr_surface = &swr->secondary;
1026       fmt = swr_surface->format;
1027    }
1028 
1029    if (rt->xpBaseAddress == swr_surface->xpBaseAddress &&
1030        rt->format == fmt &&
1031        rt->lod == sf->u.tex.level &&
1032        rt->arrayIndex == sf->u.tex.first_layer)
1033       return false;
1034 
1035    bool need_fence = false;
1036 
1037    /* StoreTile for changed target */
1038    if (rt->xpBaseAddress) {
1039       /* If changing attachment to a new target, mark tiles as
1040        * INVALID so they are reloaded from surface. */
1041       swr_store_render_target(&ctx->pipe, attachment, SWR_TILE_INVALID);
1042       need_fence = true;
1043    } else {
1044       /* if no previous attachment, invalidate tiles that may be marked
1045        * RESOLVED because of an old attachment */
1046       swr_invalidate_render_target(&ctx->pipe, attachment, sf->width, sf->height);
1047       /* no need to set fence here */
1048    }
1049 
1050    /* Make new attachment */
1051    *rt = *swr_surface;
1052    rt->format = fmt;
1053    rt->lod = sf->u.tex.level;
1054    rt->arrayIndex = sf->u.tex.first_layer;
1055 
1056    return need_fence;
1057 }
1058 
1059 /*
1060  * for cases where resources are shared between contexts, invalidate
1061  * this ctx's resource. so it can be fetched fresh.  Old ctx's resource
1062  * is already stored during a flush
1063  */
1064 static inline void
swr_invalidate_buffers_after_ctx_change(struct pipe_context * pipe)1065 swr_invalidate_buffers_after_ctx_change(struct pipe_context *pipe)
1066 {
1067    struct swr_context *ctx = swr_context(pipe);
1068 
1069    for (uint32_t i = 0; i < ctx->framebuffer.nr_cbufs; i++) {
1070       struct pipe_surface *cb = ctx->framebuffer.cbufs[i];
1071       if (cb) {
1072          struct swr_resource *res = swr_resource(cb->texture);
1073          if (res->curr_pipe != pipe) {
1074             /* if curr_pipe is NULL (first use), status should not be WRITE */
1075             assert(res->curr_pipe || !(res->status & SWR_RESOURCE_WRITE));
1076             if (res->status & SWR_RESOURCE_WRITE) {
1077                swr_invalidate_render_target(pipe, i, cb->width, cb->height);
1078             }
1079          }
1080          res->curr_pipe = pipe;
1081       }
1082    }
1083    if (ctx->framebuffer.zsbuf) {
1084       struct pipe_surface *zb = ctx->framebuffer.zsbuf;
1085       if (zb) {
1086          struct swr_resource *res = swr_resource(zb->texture);
1087          if (res->curr_pipe != pipe) {
1088             /* if curr_pipe is NULL (first use), status should not be WRITE */
1089             assert(res->curr_pipe || !(res->status & SWR_RESOURCE_WRITE));
1090             if (res->status & SWR_RESOURCE_WRITE) {
1091                swr_invalidate_render_target(pipe, SWR_ATTACHMENT_DEPTH, zb->width, zb->height);
1092                swr_invalidate_render_target(pipe, SWR_ATTACHMENT_STENCIL, zb->width, zb->height);
1093             }
1094          }
1095          res->curr_pipe = pipe;
1096       }
1097    }
1098 }
1099 
1100 static inline void
swr_user_vbuf_range(const struct pipe_draw_info * info,const struct swr_vertex_element_state * velems,const struct pipe_vertex_buffer * vb,uint32_t i,uint32_t * totelems,uint32_t * base,uint32_t * size)1101 swr_user_vbuf_range(const struct pipe_draw_info *info,
1102                     const struct swr_vertex_element_state *velems,
1103                     const struct pipe_vertex_buffer *vb,
1104                     uint32_t i,
1105                     uint32_t *totelems,
1106                     uint32_t *base,
1107                     uint32_t *size)
1108 {
1109    /* FIXME: The size is too large - we don't access the full extra stride. */
1110    unsigned elems;
1111    unsigned elem_pitch = vb->stride + velems->stream_pitch[i];
1112    if (velems->instanced_bufs & (1U << i)) {
1113       elems = info->instance_count / velems->min_instance_div[i] + 1;
1114       *totelems = info->start_instance + elems;
1115       *base = info->start_instance * vb->stride;
1116       *size = elems * elem_pitch;
1117    } else if (vb->stride) {
1118       elems = info->max_index - info->min_index + 1;
1119       *totelems = (info->max_index + info->index_bias) + 1;
1120       *base = (info->min_index + info->index_bias) * vb->stride;
1121       *size = elems * elem_pitch;
1122    } else {
1123       *totelems = 1;
1124       *base = 0;
1125       *size = velems->stream_pitch[i];
1126    }
1127 }
1128 
1129 static void
swr_update_poly_stipple(struct swr_context * ctx)1130 swr_update_poly_stipple(struct swr_context *ctx)
1131 {
1132    struct swr_draw_context *pDC = &ctx->swrDC;
1133 
1134    assert(sizeof(ctx->poly_stipple.pipe.stipple) == sizeof(pDC->polyStipple));
1135    memcpy(pDC->polyStipple,
1136           ctx->poly_stipple.pipe.stipple,
1137           sizeof(ctx->poly_stipple.pipe.stipple));
1138 }
1139 
1140 
1141 static struct tgsi_shader_info *
swr_get_last_fe(const struct swr_context * ctx)1142 swr_get_last_fe(const struct swr_context *ctx)
1143 {
1144    tgsi_shader_info *pLastFE = &ctx->vs->info.base;
1145 
1146    if (ctx->gs) {
1147       pLastFE = &ctx->gs->info.base;
1148    }
1149    else if (ctx->tes) {
1150       pLastFE = &ctx->tes->info.base;
1151    }
1152    else if (ctx->tcs) {
1153       pLastFE = &ctx->tcs->info.base;
1154    }
1155    return pLastFE;
1156 }
1157 
1158 
1159 void
swr_update_derived(struct pipe_context * pipe,const struct pipe_draw_info * p_draw_info)1160 swr_update_derived(struct pipe_context *pipe,
1161                    const struct pipe_draw_info *p_draw_info)
1162 {
1163    struct swr_context *ctx = swr_context(pipe);
1164    struct swr_screen *screen = swr_screen(pipe->screen);
1165 
1166    /* When called from swr_clear (p_draw_info = null), set any null
1167     * state-objects to the dummy state objects to prevent nullptr dereference
1168     * in validation below.
1169     *
1170     * Important that this remains static for zero initialization.  These
1171     * aren't meant to be proper state objects, just empty structs. They will
1172     * not be written to.
1173     *
1174     * Shaders can't be part of the union since they contain std::unordered_map
1175     */
1176    static struct {
1177       union {
1178          struct pipe_rasterizer_state rasterizer;
1179          struct pipe_depth_stencil_alpha_state depth_stencil;
1180          struct swr_blend_state blend;
1181       } state;
1182       struct swr_vertex_shader vs;
1183       struct swr_fragment_shader fs;
1184    } swr_dummy;
1185 
1186    if (!p_draw_info) {
1187       if (!ctx->rasterizer)
1188          ctx->rasterizer = &swr_dummy.state.rasterizer;
1189       if (!ctx->depth_stencil)
1190          ctx->depth_stencil = &swr_dummy.state.depth_stencil;
1191       if (!ctx->blend)
1192          ctx->blend = &swr_dummy.state.blend;
1193       if (!ctx->vs)
1194          ctx->vs = &swr_dummy.vs;
1195       if (!ctx->fs)
1196          ctx->fs = &swr_dummy.fs;
1197    }
1198 
1199    /* Update screen->pipe to current pipe context. */
1200    screen->pipe = pipe;
1201 
1202    /* Any state that requires dirty flags to be re-triggered sets this mask */
1203    /* For example, user_buffer vertex and index buffers. */
1204    unsigned post_update_dirty_flags = 0;
1205 
1206    /* bring resources that changed context up-to-date */
1207    swr_invalidate_buffers_after_ctx_change(pipe);
1208 
1209    /* Render Targets */
1210    if (ctx->dirty & SWR_NEW_FRAMEBUFFER) {
1211       struct pipe_framebuffer_state *fb = &ctx->framebuffer;
1212       const struct util_format_description *desc = NULL;
1213       bool need_fence = false;
1214 
1215       /* colorbuffer targets */
1216       if (fb->nr_cbufs) {
1217          for (unsigned i = 0; i < fb->nr_cbufs; ++i)
1218             need_fence |= swr_change_rt(
1219                   ctx, SWR_ATTACHMENT_COLOR0 + i, fb->cbufs[i]);
1220       }
1221       for (unsigned i = fb->nr_cbufs; i < SWR_NUM_RENDERTARGETS; ++i)
1222          need_fence |= swr_change_rt(ctx, SWR_ATTACHMENT_COLOR0 + i, NULL);
1223 
1224       /* depth/stencil target */
1225       if (fb->zsbuf)
1226          desc = util_format_description(fb->zsbuf->format);
1227       if (fb->zsbuf && util_format_has_depth(desc))
1228          need_fence |= swr_change_rt(ctx, SWR_ATTACHMENT_DEPTH, fb->zsbuf);
1229       else
1230          need_fence |= swr_change_rt(ctx, SWR_ATTACHMENT_DEPTH, NULL);
1231 
1232       if (fb->zsbuf && util_format_has_stencil(desc))
1233          need_fence |= swr_change_rt(ctx, SWR_ATTACHMENT_STENCIL, fb->zsbuf);
1234       else
1235          need_fence |= swr_change_rt(ctx, SWR_ATTACHMENT_STENCIL, NULL);
1236 
1237       /* This fence ensures any attachment changes are resolved before the
1238        * next draw */
1239       if (need_fence)
1240          swr_fence_submit(ctx, screen->flush_fence);
1241    }
1242 
1243    /* Raster state */
1244    if (ctx->dirty & (SWR_NEW_RASTERIZER |
1245                      SWR_NEW_VS | // clipping
1246                      SWR_NEW_TES |
1247                      SWR_NEW_TCS |
1248                      SWR_NEW_FRAMEBUFFER)) {
1249       pipe_rasterizer_state *rasterizer = ctx->rasterizer;
1250       pipe_framebuffer_state *fb = &ctx->framebuffer;
1251 
1252       SWR_RASTSTATE *rastState = &ctx->derived.rastState;
1253       rastState->cullMode = swr_convert_cull_mode(rasterizer->cull_face);
1254       rastState->frontWinding = rasterizer->front_ccw
1255          ? SWR_FRONTWINDING_CCW
1256          : SWR_FRONTWINDING_CW;
1257       rastState->scissorEnable = rasterizer->scissor;
1258       rastState->pointSize = rasterizer->point_size > 0.0f
1259          ? rasterizer->point_size
1260          : 1.0f;
1261       rastState->lineWidth = rasterizer->line_width > 0.0f
1262          ? rasterizer->line_width
1263          : 1.0f;
1264 
1265       rastState->pointParam = rasterizer->point_size_per_vertex;
1266 
1267       rastState->pointSpriteEnable = rasterizer->sprite_coord_enable;
1268       rastState->pointSpriteTopOrigin =
1269          rasterizer->sprite_coord_mode == PIPE_SPRITE_COORD_UPPER_LEFT;
1270 
1271       /* If SWR_MSAA_FORCE_ENABLE is set, turn msaa on */
1272       if (screen->msaa_force_enable && !rasterizer->multisample) {
1273          /* Force enable and use the value the surface was created with */
1274          rasterizer->multisample = true;
1275          fb->samples = swr_resource(fb->cbufs[0]->texture)->swr.numSamples;
1276          fprintf(stderr,"msaa force enable: %d samples\n", fb->samples);
1277       }
1278 
1279       rastState->sampleCount = GetSampleCount(fb->samples);
1280       rastState->forcedSampleCount = false;
1281       rastState->bIsCenterPattern = !rasterizer->multisample;
1282       rastState->pixelLocation = SWR_PIXEL_LOCATION_CENTER;
1283 
1284       /* Only initialize sample positions if msaa is enabled */
1285       if (rasterizer->multisample) {
1286          for (uint32_t i = 0; i < fb->samples; i++) {
1287             const uint8_t *sample = swr_sample_positions[fb->samples-1 + i];
1288             rastState->samplePositions.SetXi(i, sample[0] << 4);
1289             rastState->samplePositions.SetYi(i, sample[1] << 4);
1290             rastState->samplePositions.SetX (i, sample[0] / 16.0f);
1291             rastState->samplePositions.SetY (i, sample[1] / 16.0f);
1292          }
1293          rastState->samplePositions.PrecalcSampleData(fb->samples);
1294       }
1295 
1296       bool do_offset = false;
1297       switch (rasterizer->fill_front) {
1298       case PIPE_POLYGON_MODE_FILL:
1299          do_offset = rasterizer->offset_tri;
1300          break;
1301       case PIPE_POLYGON_MODE_LINE:
1302          do_offset = rasterizer->offset_line;
1303          break;
1304       case PIPE_POLYGON_MODE_POINT:
1305          do_offset = rasterizer->offset_point;
1306          break;
1307       }
1308 
1309       if (do_offset) {
1310          rastState->depthBias = rasterizer->offset_units;
1311          rastState->slopeScaledDepthBias = rasterizer->offset_scale;
1312          rastState->depthBiasClamp = rasterizer->offset_clamp;
1313       } else {
1314          rastState->depthBias = 0;
1315          rastState->slopeScaledDepthBias = 0;
1316          rastState->depthBiasClamp = 0;
1317       }
1318 
1319       /* translate polygon mode, at least for the front==back case */
1320       rastState->fillMode = swr_convert_fill_mode(rasterizer->fill_front);
1321 
1322       struct pipe_surface *zb = fb->zsbuf;
1323       if (zb && swr_resource(zb->texture)->has_depth)
1324          rastState->depthFormat = swr_resource(zb->texture)->swr.format;
1325 
1326       rastState->depthClipEnable = rasterizer->depth_clip_near;
1327       rastState->clipEnable = rasterizer->depth_clip_near | rasterizer->depth_clip_far;
1328       rastState->clipHalfZ = rasterizer->clip_halfz;
1329 
1330       ctx->api.pfnSwrSetRastState(ctx->swrContext, rastState);
1331    }
1332 
1333    /* Viewport */
1334    if (ctx->dirty & (SWR_NEW_VIEWPORT | SWR_NEW_FRAMEBUFFER
1335                      | SWR_NEW_RASTERIZER)) {
1336       pipe_viewport_state *state = &ctx->viewports[0];
1337       pipe_framebuffer_state *fb = &ctx->framebuffer;
1338       pipe_rasterizer_state *rasterizer = ctx->rasterizer;
1339 
1340       SWR_VIEWPORT *vp = &ctx->derived.vp[0];
1341       SWR_VIEWPORT_MATRICES *vpm = &ctx->derived.vpm;
1342 
1343       for (unsigned i = 0; i < KNOB_NUM_VIEWPORTS_SCISSORS; i++) {
1344          vp->x = state->translate[0] - state->scale[0];
1345          vp->width = 2 * state->scale[0];
1346          vp->y = state->translate[1] - fabs(state->scale[1]);
1347          vp->height = 2 * fabs(state->scale[1]);
1348          util_viewport_zmin_zmax(state, rasterizer->clip_halfz,
1349                                  &vp->minZ, &vp->maxZ);
1350 
1351          if (rasterizer->depth_clip_near) {
1352             vp->minZ = 0.0f;
1353          }
1354 
1355          if (rasterizer->depth_clip_far) {
1356             vp->maxZ = 1.0f;
1357          }
1358 
1359          vpm->m00[i] = state->scale[0];
1360          vpm->m11[i] = state->scale[1];
1361          vpm->m22[i] = state->scale[2];
1362          vpm->m30[i] = state->translate[0];
1363          vpm->m31[i] = state->translate[1];
1364          vpm->m32[i] = state->translate[2];
1365 
1366          /* Now that the matrix is calculated, clip the view coords to screen
1367           * size.  OpenGL allows for -ve x,y in the viewport. */
1368          if (vp->x < 0.0f) {
1369             vp->width += vp->x;
1370             vp->x = 0.0f;
1371          }
1372          if (vp->y < 0.0f) {
1373             vp->height += vp->y;
1374             vp->y = 0.0f;
1375          }
1376          vp->width = std::min(vp->width, (float) fb->width - vp->x);
1377          vp->height = std::min(vp->height, (float) fb->height - vp->y);
1378 
1379          vp++;
1380          state++;
1381       }
1382       ctx->api.pfnSwrSetViewports(ctx->swrContext, KNOB_NUM_VIEWPORTS_SCISSORS,
1383                                   &ctx->derived.vp[0], &ctx->derived.vpm);
1384    }
1385 
1386    /* When called from swr_clear (p_draw_info = null), render targets,
1387     * rasterState and viewports (dependent on render targets) are the only
1388     * necessary validation.  Defer remaining validation by setting
1389     * post_update_dirty_flags and clear all dirty flags.  BackendState is
1390     * still unconditionally validated below */
1391    if (!p_draw_info) {
1392       post_update_dirty_flags = ctx->dirty & ~(SWR_NEW_FRAMEBUFFER |
1393                                                SWR_NEW_RASTERIZER |
1394                                                SWR_NEW_VIEWPORT);
1395       ctx->dirty = 0;
1396    }
1397 
1398    /* Scissor */
1399    if (ctx->dirty & SWR_NEW_SCISSOR) {
1400       ctx->api.pfnSwrSetScissorRects(ctx->swrContext, KNOB_NUM_VIEWPORTS_SCISSORS, ctx->swr_scissors);
1401    }
1402 
1403    /* Set vertex & index buffers */
1404    if (ctx->dirty & SWR_NEW_VERTEX) {
1405       const struct pipe_draw_info &info = *p_draw_info;
1406 
1407       /* vertex buffers */
1408       SWR_VERTEX_BUFFER_STATE swrVertexBuffers[PIPE_MAX_ATTRIBS];
1409       for (UINT i = 0; i < ctx->num_vertex_buffers; i++) {
1410          uint32_t size = 0, pitch = 0, elems = 0, partial_inbounds = 0;
1411          uint32_t min_vertex_index = 0;
1412          const uint8_t *p_data;
1413          struct pipe_vertex_buffer *vb = &ctx->vertex_buffer[i];
1414 
1415          pitch = vb->stride;
1416          if (vb->is_user_buffer) {
1417             /* Client buffer
1418              * client memory is one-time use, re-trigger SWR_NEW_VERTEX to
1419              * revalidate on each draw */
1420             post_update_dirty_flags |= SWR_NEW_VERTEX;
1421 
1422             uint32_t base;
1423             swr_user_vbuf_range(&info, ctx->velems, vb, i, &elems, &base, &size);
1424             partial_inbounds = 0;
1425             min_vertex_index = info.min_index + info.index_bias;
1426 
1427             size = AlignUp(size, 4);
1428             /* If size of client memory copy is too large, don't copy. The
1429              * draw will access user-buffer directly and then block.  This is
1430              * faster than queuing many large client draws. */
1431             if (size >= screen->client_copy_limit) {
1432                post_update_dirty_flags |= SWR_BLOCK_CLIENT_DRAW;
1433                p_data = (const uint8_t *) vb->buffer.user;
1434             } else {
1435                /* Copy only needed vertices to scratch space */
1436                const void *ptr = (const uint8_t *) vb->buffer.user + base;
1437                ptr = (uint8_t *)swr_copy_to_scratch_space(
1438                      ctx, &ctx->scratch->vertex_buffer, ptr, size);
1439                p_data = (const uint8_t *)ptr - base;
1440             }
1441          } else if (vb->buffer.resource) {
1442             /* VBO */
1443             if (!pitch) {
1444                /* If pitch=0 (ie vb->stride), buffer contains a single
1445                 * constant attribute.  Use the stream_pitch which was
1446                 * calculated during creation of vertex_elements_state for the
1447                 * size of the attribute. */
1448                size = ctx->velems->stream_pitch[i];
1449                elems = 1;
1450                partial_inbounds = 0;
1451                min_vertex_index = 0;
1452             } else {
1453                /* size is based on buffer->width0 rather than info.max_index
1454                 * to prevent having to validate VBO on each draw. */
1455                size = vb->buffer.resource->width0;
1456                elems = size / pitch;
1457                partial_inbounds = size % pitch;
1458                min_vertex_index = 0;
1459             }
1460 
1461             p_data = swr_resource_data(vb->buffer.resource) + vb->buffer_offset;
1462          } else
1463             p_data = NULL;
1464 
1465          swrVertexBuffers[i] = {0};
1466          swrVertexBuffers[i].index = i;
1467          swrVertexBuffers[i].pitch = pitch;
1468          swrVertexBuffers[i].xpData = (gfxptr_t) p_data;
1469          swrVertexBuffers[i].size = size;
1470          swrVertexBuffers[i].minVertex = min_vertex_index;
1471          swrVertexBuffers[i].maxVertex = elems;
1472          swrVertexBuffers[i].partialInboundsSize = partial_inbounds;
1473       }
1474 
1475       ctx->api.pfnSwrSetVertexBuffers(
1476          ctx->swrContext, ctx->num_vertex_buffers, swrVertexBuffers);
1477 
1478       /* index buffer, if required (info passed in by swr_draw_vbo) */
1479       SWR_FORMAT index_type = R32_UINT; /* Default for non-indexed draws */
1480       if (info.index_size) {
1481          const uint8_t *p_data;
1482          uint32_t size, pitch;
1483 
1484          pitch = info.index_size ? info.index_size : sizeof(uint32_t);
1485          index_type = swr_convert_index_type(pitch);
1486 
1487          if (!info.has_user_indices) {
1488             /* VBO
1489              * size is based on buffer->width0 rather than info.count
1490              * to prevent having to validate VBO on each draw */
1491             size = info.index.resource->width0;
1492             p_data = swr_resource_data(info.index.resource);
1493          } else {
1494             /* Client buffer
1495              * client memory is one-time use, re-trigger SWR_NEW_VERTEX to
1496              * revalidate on each draw */
1497             post_update_dirty_flags |= SWR_NEW_VERTEX;
1498 
1499             size = info.count * pitch;
1500 
1501             size = AlignUp(size, 4);
1502             /* If size of client memory copy is too large, don't copy. The
1503              * draw will access user-buffer directly and then block.  This is
1504              * faster than queuing many large client draws. */
1505             if (size >= screen->client_copy_limit) {
1506                post_update_dirty_flags |= SWR_BLOCK_CLIENT_DRAW;
1507                p_data = (const uint8_t *) info.index.user;
1508             } else {
1509                /* Copy indices to scratch space */
1510                const void *ptr = info.index.user;
1511                ptr = swr_copy_to_scratch_space(
1512                      ctx, &ctx->scratch->index_buffer, ptr, size);
1513                p_data = (const uint8_t *)ptr;
1514             }
1515          }
1516 
1517          SWR_INDEX_BUFFER_STATE swrIndexBuffer;
1518          swrIndexBuffer.format = swr_convert_index_type(info.index_size);
1519          swrIndexBuffer.xpIndices = (gfxptr_t) p_data;
1520          swrIndexBuffer.size = size;
1521 
1522          ctx->api.pfnSwrSetIndexBuffer(ctx->swrContext, &swrIndexBuffer);
1523       }
1524 
1525       struct swr_vertex_element_state *velems = ctx->velems;
1526       if (velems && velems->fsState.indexType != index_type) {
1527          velems->fsFunc = NULL;
1528          velems->fsState.indexType = index_type;
1529       }
1530    }
1531 
1532    /* GeometryShader */
1533    if (ctx->dirty & (SWR_NEW_GS |
1534                      SWR_NEW_VS |
1535                      SWR_NEW_TCS |
1536                      SWR_NEW_TES |
1537                      SWR_NEW_SAMPLER |
1538                      SWR_NEW_SAMPLER_VIEW)) {
1539       if (ctx->gs) {
1540          swr_jit_gs_key key;
1541          swr_generate_gs_key(key, ctx, ctx->gs);
1542          auto search = ctx->gs->map.find(key);
1543          PFN_GS_FUNC func;
1544          if (search != ctx->gs->map.end()) {
1545             func = search->second->shader;
1546          } else {
1547             func = swr_compile_gs(ctx, key);
1548          }
1549          ctx->api.pfnSwrSetGsFunc(ctx->swrContext, func);
1550 
1551          /* JIT sampler state */
1552          if (ctx->dirty & SWR_NEW_SAMPLER) {
1553             swr_update_sampler_state(ctx,
1554                                      PIPE_SHADER_GEOMETRY,
1555                                      key.nr_samplers,
1556                                      ctx->swrDC.samplersGS);
1557          }
1558 
1559          /* JIT sampler view state */
1560          if (ctx->dirty & (SWR_NEW_SAMPLER_VIEW | SWR_NEW_FRAMEBUFFER)) {
1561             swr_update_texture_state(ctx,
1562                                      PIPE_SHADER_GEOMETRY,
1563                                      key.nr_sampler_views,
1564                                      ctx->swrDC.texturesGS);
1565          }
1566 
1567          ctx->api.pfnSwrSetGsState(ctx->swrContext, &ctx->gs->gsState);
1568       } else {
1569          SWR_GS_STATE state = { 0 };
1570          ctx->api.pfnSwrSetGsState(ctx->swrContext, &state);
1571          ctx->api.pfnSwrSetGsFunc(ctx->swrContext, NULL);
1572       }
1573    }
1574 
1575    // We may need to restore tessellation state
1576    // This restored state may be however overwritten
1577    // during shader compilation
1578    if (ctx->dirty & SWR_NEW_TS) {
1579       if (ctx->tes != nullptr) {
1580          ctx->tsState = ctx->tes->ts_state;
1581          ctx->api.pfnSwrSetTsState(ctx->swrContext, &ctx->tsState);
1582       } else {
1583          SWR_TS_STATE state = { 0 };
1584          ctx->api.pfnSwrSetTsState(ctx->swrContext, &state);
1585       }
1586    }
1587 
1588    // Tessellation Evaluation Shader
1589    // Compile TES first, because TCS is optional
1590    if (ctx->dirty & (SWR_NEW_GS |
1591                      SWR_NEW_VS |
1592                      SWR_NEW_TCS |
1593                      SWR_NEW_TES |
1594                      SWR_NEW_SAMPLER |
1595                      SWR_NEW_SAMPLER_VIEW)) {
1596       if (ctx->tes) {
1597          swr_jit_tes_key key;
1598          swr_generate_tes_key(key, ctx, ctx->tes);
1599 
1600          auto search = ctx->tes->map.find(key);
1601          PFN_TES_FUNC func;
1602          if (search != ctx->tes->map.end()) {
1603             func = search->second->shader;
1604          } else {
1605             func = swr_compile_tes(ctx, key);
1606          }
1607 
1608          ctx->api.pfnSwrSetDsFunc(ctx->swrContext, func);
1609 
1610          /* JIT sampler state */
1611          if (ctx->dirty & SWR_NEW_SAMPLER) {
1612             swr_update_sampler_state(ctx,
1613                                      PIPE_SHADER_TESS_EVAL,
1614                                      key.nr_samplers,
1615                                      ctx->swrDC.samplersTES);
1616          }
1617 
1618          /* JIT sampler view state */
1619          if (ctx->dirty & (SWR_NEW_SAMPLER_VIEW | SWR_NEW_FRAMEBUFFER)) {
1620             swr_update_texture_state(ctx,
1621                                      PIPE_SHADER_TESS_EVAL,
1622                                      key.nr_sampler_views,
1623                                      ctx->swrDC.texturesTES);
1624          }
1625 
1626          // Update tessellation state in case it's been updated
1627          ctx->api.pfnSwrSetTsState(ctx->swrContext, &ctx->tsState);
1628       } else {
1629          ctx->api.pfnSwrSetDsFunc(ctx->swrContext, NULL);
1630       }
1631    }
1632 
1633    /* Tessellation Control Shader */
1634    if (ctx->dirty & (SWR_NEW_GS |
1635                      SWR_NEW_VS |
1636                      SWR_NEW_TCS |
1637                      SWR_NEW_TES |
1638                      SWR_NEW_SAMPLER |
1639                      SWR_NEW_SAMPLER_VIEW)) {
1640       if (ctx->tcs) {
1641          ctx->tcs->vertices_per_patch = p_draw_info->vertices_per_patch;
1642 
1643          swr_jit_tcs_key key;
1644          swr_generate_tcs_key(key, ctx, ctx->tcs);
1645 
1646          auto search = ctx->tcs->map.find(key);
1647          PFN_TCS_FUNC func;
1648          if (search != ctx->tcs->map.end()) {
1649             func = search->second->shader;
1650          } else {
1651             func = swr_compile_tcs(ctx, key);
1652          }
1653 
1654          ctx->api.pfnSwrSetHsFunc(ctx->swrContext, func);
1655 
1656          /* JIT sampler state */
1657          if (ctx->dirty & SWR_NEW_SAMPLER) {
1658             swr_update_sampler_state(ctx,
1659                                      PIPE_SHADER_TESS_CTRL,
1660                                      key.nr_samplers,
1661                                      ctx->swrDC.samplersTCS);
1662          }
1663 
1664          /* JIT sampler view state */
1665          if (ctx->dirty & (SWR_NEW_SAMPLER_VIEW | SWR_NEW_FRAMEBUFFER)) {
1666             swr_update_texture_state(ctx,
1667                                      PIPE_SHADER_TESS_CTRL,
1668                                      key.nr_sampler_views,
1669                                      ctx->swrDC.texturesTCS);
1670          }
1671 
1672          // Update tessellation state in case it's been updated
1673          ctx->api.pfnSwrSetTsState(ctx->swrContext, &ctx->tsState);
1674       } else {
1675          ctx->api.pfnSwrSetHsFunc(ctx->swrContext, NULL);
1676       }
1677    }
1678 
1679    /* VertexShader */
1680    if (ctx->dirty
1681        & (SWR_NEW_VS | SWR_NEW_RASTERIZER | // for clip planes
1682           SWR_NEW_SAMPLER | SWR_NEW_SAMPLER_VIEW | SWR_NEW_FRAMEBUFFER)) {
1683       swr_jit_vs_key key;
1684       swr_generate_vs_key(key, ctx, ctx->vs);
1685       auto search = ctx->vs->map.find(key);
1686       PFN_VERTEX_FUNC func;
1687       if (search != ctx->vs->map.end()) {
1688          func = search->second->shader;
1689       } else {
1690          func = swr_compile_vs(ctx, key);
1691       }
1692       ctx->api.pfnSwrSetVertexFunc(ctx->swrContext, func);
1693 
1694       /* JIT sampler state */
1695       if (ctx->dirty & SWR_NEW_SAMPLER) {
1696          swr_update_sampler_state(
1697             ctx, PIPE_SHADER_VERTEX, key.nr_samplers, ctx->swrDC.samplersVS);
1698       }
1699 
1700       /* JIT sampler view state */
1701       if (ctx->dirty & (SWR_NEW_SAMPLER_VIEW | SWR_NEW_FRAMEBUFFER)) {
1702          swr_update_texture_state(ctx,
1703                                   PIPE_SHADER_VERTEX,
1704                                   key.nr_sampler_views,
1705                                   ctx->swrDC.texturesVS);
1706       }
1707    }
1708 
1709    /* work around the fact that poly stipple also affects lines */
1710    /* and points, since we rasterize them as triangles, too */
1711    /* Has to be before fragment shader, since it sets SWR_NEW_FS */
1712    if (p_draw_info) {
1713       bool new_prim_is_poly =
1714          (u_reduced_prim(p_draw_info->mode) == PIPE_PRIM_TRIANGLES) &&
1715          (ctx->derived.rastState.fillMode == SWR_FILLMODE_SOLID);
1716       if (new_prim_is_poly != ctx->poly_stipple.prim_is_poly) {
1717          ctx->dirty |= SWR_NEW_FS;
1718          ctx->poly_stipple.prim_is_poly = new_prim_is_poly;
1719       }
1720    }
1721 
1722    /* FragmentShader */
1723    if (ctx->dirty & (SWR_NEW_FS |
1724                      SWR_NEW_VS |
1725                      SWR_NEW_GS |
1726                      SWR_NEW_TES |
1727                      SWR_NEW_TCS |
1728                      SWR_NEW_RASTERIZER |
1729                      SWR_NEW_SAMPLER |
1730                      SWR_NEW_SAMPLER_VIEW |
1731                      SWR_NEW_FRAMEBUFFER)) {
1732       swr_jit_fs_key key;
1733       swr_generate_fs_key(key, ctx, ctx->fs);
1734       auto search = ctx->fs->map.find(key);
1735       PFN_PIXEL_KERNEL func;
1736       if (search != ctx->fs->map.end()) {
1737          func = search->second->shader;
1738       } else {
1739          func = swr_compile_fs(ctx, key);
1740       }
1741       SWR_PS_STATE psState = {0};
1742       psState.pfnPixelShader = func;
1743       psState.killsPixel = ctx->fs->info.base.uses_kill;
1744       psState.inputCoverage = SWR_INPUT_COVERAGE_NORMAL;
1745       psState.writesODepth = ctx->fs->info.base.writes_z;
1746       psState.usesSourceDepth = ctx->fs->info.base.reads_z;
1747       psState.shadingRate = SWR_SHADING_RATE_PIXEL;
1748       psState.renderTargetMask = (1 << ctx->framebuffer.nr_cbufs) - 1;
1749       psState.posOffset = SWR_PS_POSITION_SAMPLE_NONE;
1750       uint32_t barycentricsMask = 0;
1751 #if 0
1752       // when we switch to mesa-master
1753       if (ctx->fs->info.base.uses_persp_center ||
1754           ctx->fs->info.base.uses_linear_center)
1755          barycentricsMask |= SWR_BARYCENTRIC_PER_PIXEL_MASK;
1756       if (ctx->fs->info.base.uses_persp_centroid ||
1757           ctx->fs->info.base.uses_linear_centroid)
1758          barycentricsMask |= SWR_BARYCENTRIC_CENTROID_MASK;
1759       if (ctx->fs->info.base.uses_persp_sample ||
1760           ctx->fs->info.base.uses_linear_sample)
1761          barycentricsMask |= SWR_BARYCENTRIC_PER_SAMPLE_MASK;
1762 #else
1763       for (unsigned i = 0; i < ctx->fs->info.base.num_inputs; i++) {
1764          switch (ctx->fs->info.base.input_interpolate_loc[i]) {
1765          case TGSI_INTERPOLATE_LOC_CENTER:
1766             barycentricsMask |= SWR_BARYCENTRIC_PER_PIXEL_MASK;
1767             break;
1768          case TGSI_INTERPOLATE_LOC_CENTROID:
1769             barycentricsMask |= SWR_BARYCENTRIC_CENTROID_MASK;
1770             break;
1771          case TGSI_INTERPOLATE_LOC_SAMPLE:
1772             barycentricsMask |= SWR_BARYCENTRIC_PER_SAMPLE_MASK;
1773             break;
1774          }
1775       }
1776 #endif
1777       psState.barycentricsMask = barycentricsMask;
1778       psState.usesUAV = false; // XXX
1779       psState.forceEarlyZ = false;
1780       ctx->api.pfnSwrSetPixelShaderState(ctx->swrContext, &psState);
1781 
1782       /* JIT sampler state */
1783       if (ctx->dirty & (SWR_NEW_SAMPLER |
1784                         SWR_NEW_FS)) {
1785          swr_update_sampler_state(ctx,
1786                                   PIPE_SHADER_FRAGMENT,
1787                                   key.nr_samplers,
1788                                   ctx->swrDC.samplersFS);
1789       }
1790 
1791       /* JIT sampler view state */
1792       if (ctx->dirty & (SWR_NEW_SAMPLER_VIEW |
1793                         SWR_NEW_FRAMEBUFFER |
1794                         SWR_NEW_FS)) {
1795          swr_update_texture_state(ctx,
1796                                   PIPE_SHADER_FRAGMENT,
1797                                   key.nr_sampler_views,
1798                                   ctx->swrDC.texturesFS);
1799       }
1800    }
1801 
1802 
1803    /* VertexShader Constants */
1804    if (ctx->dirty & SWR_NEW_VSCONSTANTS) {
1805       swr_update_constants(ctx, PIPE_SHADER_VERTEX);
1806    }
1807 
1808    /* FragmentShader Constants */
1809    if (ctx->dirty & SWR_NEW_FSCONSTANTS) {
1810       swr_update_constants(ctx, PIPE_SHADER_FRAGMENT);
1811    }
1812 
1813    /* GeometryShader Constants */
1814    if (ctx->dirty & SWR_NEW_GSCONSTANTS) {
1815       swr_update_constants(ctx, PIPE_SHADER_GEOMETRY);
1816    }
1817 
1818    /* Tessellation Control Shader Constants */
1819    if (ctx->dirty & SWR_NEW_TCSCONSTANTS) {
1820       swr_update_constants(ctx, PIPE_SHADER_TESS_CTRL);
1821    }
1822 
1823    /* Tessellation Evaluation Shader Constants */
1824    if (ctx->dirty & SWR_NEW_TESCONSTANTS) {
1825       swr_update_constants(ctx, PIPE_SHADER_TESS_EVAL);
1826    }
1827 
1828    /* Depth/stencil state */
1829    if (ctx->dirty & (SWR_NEW_DEPTH_STENCIL_ALPHA | SWR_NEW_FRAMEBUFFER)) {
1830       struct pipe_depth_state *depth = &(ctx->depth_stencil->depth);
1831       struct pipe_stencil_state *stencil = ctx->depth_stencil->stencil;
1832       SWR_DEPTH_STENCIL_STATE depthStencilState = {{0}};
1833       SWR_DEPTH_BOUNDS_STATE depthBoundsState = {0};
1834 
1835       /* XXX, incomplete.  Need to flesh out stencil & alpha test state
1836       struct pipe_stencil_state *front_stencil =
1837       ctx->depth_stencil.stencil[0];
1838       struct pipe_stencil_state *back_stencil = ctx->depth_stencil.stencil[1];
1839       struct pipe_alpha_state alpha;
1840       */
1841       if (stencil[0].enabled) {
1842          depthStencilState.stencilWriteEnable = 1;
1843          depthStencilState.stencilTestEnable = 1;
1844          depthStencilState.stencilTestFunc =
1845             swr_convert_depth_func(stencil[0].func);
1846 
1847          depthStencilState.stencilPassDepthPassOp =
1848             swr_convert_stencil_op(stencil[0].zpass_op);
1849          depthStencilState.stencilPassDepthFailOp =
1850             swr_convert_stencil_op(stencil[0].zfail_op);
1851          depthStencilState.stencilFailOp =
1852             swr_convert_stencil_op(stencil[0].fail_op);
1853          depthStencilState.stencilWriteMask = stencil[0].writemask;
1854          depthStencilState.stencilTestMask = stencil[0].valuemask;
1855          depthStencilState.stencilRefValue = ctx->stencil_ref.ref_value[0];
1856       }
1857       if (stencil[1].enabled) {
1858          depthStencilState.doubleSidedStencilTestEnable = 1;
1859 
1860          depthStencilState.backfaceStencilTestFunc =
1861             swr_convert_depth_func(stencil[1].func);
1862 
1863          depthStencilState.backfaceStencilPassDepthPassOp =
1864             swr_convert_stencil_op(stencil[1].zpass_op);
1865          depthStencilState.backfaceStencilPassDepthFailOp =
1866             swr_convert_stencil_op(stencil[1].zfail_op);
1867          depthStencilState.backfaceStencilFailOp =
1868             swr_convert_stencil_op(stencil[1].fail_op);
1869          depthStencilState.backfaceStencilWriteMask = stencil[1].writemask;
1870          depthStencilState.backfaceStencilTestMask = stencil[1].valuemask;
1871 
1872          depthStencilState.backfaceStencilRefValue =
1873             ctx->stencil_ref.ref_value[1];
1874       }
1875 
1876       depthStencilState.depthTestEnable = depth->enabled;
1877       depthStencilState.depthTestFunc = swr_convert_depth_func(depth->func);
1878       depthStencilState.depthWriteEnable = depth->writemask;
1879       ctx->api.pfnSwrSetDepthStencilState(ctx->swrContext, &depthStencilState);
1880 
1881       depthBoundsState.depthBoundsTestEnable = depth->bounds_test;
1882       depthBoundsState.depthBoundsTestMinValue = depth->bounds_min;
1883       depthBoundsState.depthBoundsTestMaxValue = depth->bounds_max;
1884       ctx->api.pfnSwrSetDepthBoundsState(ctx->swrContext, &depthBoundsState);
1885    }
1886 
1887    /* Blend State */
1888    if (ctx->dirty & (SWR_NEW_BLEND |
1889                      SWR_NEW_RASTERIZER |
1890                      SWR_NEW_FRAMEBUFFER |
1891                      SWR_NEW_DEPTH_STENCIL_ALPHA)) {
1892       struct pipe_framebuffer_state *fb = &ctx->framebuffer;
1893 
1894       SWR_BLEND_STATE blendState;
1895       memcpy(&blendState, &ctx->blend->blendState, sizeof(blendState));
1896       blendState.constantColor[0] = ctx->blend_color.color[0];
1897       blendState.constantColor[1] = ctx->blend_color.color[1];
1898       blendState.constantColor[2] = ctx->blend_color.color[2];
1899       blendState.constantColor[3] = ctx->blend_color.color[3];
1900       blendState.alphaTestReference =
1901          *((uint32_t*)&ctx->depth_stencil->alpha.ref_value);
1902 
1903       blendState.sampleMask = ctx->sample_mask;
1904       blendState.sampleCount = GetSampleCount(fb->samples);
1905 
1906       /* If there are no color buffers bound, disable writes on RT0
1907        * and skip loop */
1908       if (fb->nr_cbufs == 0) {
1909          blendState.renderTarget[0].writeDisableRed = 1;
1910          blendState.renderTarget[0].writeDisableGreen = 1;
1911          blendState.renderTarget[0].writeDisableBlue = 1;
1912          blendState.renderTarget[0].writeDisableAlpha = 1;
1913          ctx->api.pfnSwrSetBlendFunc(ctx->swrContext, 0, NULL);
1914       }
1915       else
1916          for (int target = 0;
1917                target < std::min(SWR_NUM_RENDERTARGETS,
1918                                  PIPE_MAX_COLOR_BUFS);
1919                target++) {
1920             if (!fb->cbufs[target])
1921                continue;
1922 
1923             struct swr_resource *colorBuffer =
1924                swr_resource(fb->cbufs[target]->texture);
1925 
1926             BLEND_COMPILE_STATE compileState;
1927             memset(&compileState, 0, sizeof(compileState));
1928             compileState.format = colorBuffer->swr.format;
1929             memcpy(&compileState.blendState,
1930                    &ctx->blend->compileState[target],
1931                    sizeof(compileState.blendState));
1932 
1933             const SWR_FORMAT_INFO& info = GetFormatInfo(compileState.format);
1934             if (compileState.blendState.logicOpEnable &&
1935                 ((info.type[0] == SWR_TYPE_FLOAT) || info.isSRGB)) {
1936                compileState.blendState.logicOpEnable = false;
1937             }
1938 
1939             if (info.type[0] == SWR_TYPE_SINT || info.type[0] == SWR_TYPE_UINT)
1940                compileState.blendState.blendEnable = false;
1941 
1942             if (compileState.blendState.blendEnable == false &&
1943                 compileState.blendState.logicOpEnable == false &&
1944                 ctx->depth_stencil->alpha.enabled == 0) {
1945                ctx->api.pfnSwrSetBlendFunc(ctx->swrContext, target, NULL);
1946                continue;
1947             }
1948 
1949             compileState.desc.alphaTestEnable =
1950                ctx->depth_stencil->alpha.enabled;
1951             compileState.desc.independentAlphaBlendEnable =
1952                (compileState.blendState.sourceBlendFactor !=
1953                 compileState.blendState.sourceAlphaBlendFactor) ||
1954                (compileState.blendState.destBlendFactor !=
1955                 compileState.blendState.destAlphaBlendFactor) ||
1956                (compileState.blendState.colorBlendFunc !=
1957                 compileState.blendState.alphaBlendFunc);
1958             compileState.desc.alphaToCoverageEnable =
1959                ctx->blend->pipe.alpha_to_coverage;
1960             compileState.desc.sampleMaskEnable = (blendState.sampleMask != 0);
1961             compileState.desc.numSamples = fb->samples;
1962 
1963             compileState.alphaTestFunction =
1964                swr_convert_depth_func(ctx->depth_stencil->alpha.func);
1965             compileState.alphaTestFormat = ALPHA_TEST_FLOAT32; // xxx
1966 
1967             compileState.Canonicalize();
1968 
1969             PFN_BLEND_JIT_FUNC func = NULL;
1970             auto search = ctx->blendJIT->find(compileState);
1971             if (search != ctx->blendJIT->end()) {
1972                func = search->second;
1973             } else {
1974                HANDLE hJitMgr = screen->hJitMgr;
1975                func = JitCompileBlend(hJitMgr, compileState);
1976                debug_printf("BLEND shader %p\n", func);
1977                assert(func && "Error: BlendShader = NULL");
1978 
1979                ctx->blendJIT->insert(std::make_pair(compileState, func));
1980             }
1981             ctx->api.pfnSwrSetBlendFunc(ctx->swrContext, target, func);
1982          }
1983 
1984       ctx->api.pfnSwrSetBlendState(ctx->swrContext, &blendState);
1985    }
1986 
1987    if (ctx->dirty & SWR_NEW_STIPPLE) {
1988       swr_update_poly_stipple(ctx);
1989    }
1990 
1991    if (ctx->dirty & (SWR_NEW_VS | SWR_NEW_TCS | SWR_NEW_TES | SWR_NEW_SO | SWR_NEW_RASTERIZER)) {
1992       ctx->vs->soState.rasterizerDisable =
1993          ctx->rasterizer->rasterizer_discard;
1994       ctx->api.pfnSwrSetSoState(ctx->swrContext, &ctx->vs->soState);
1995 
1996       pipe_stream_output_info *stream_output = &ctx->vs->pipe.stream_output;
1997 
1998       for (uint32_t i = 0; i < MAX_SO_STREAMS; i++) {
1999          SWR_STREAMOUT_BUFFER buffer = {0};
2000          if (ctx->so_targets[i]) {
2001              buffer.enable = true;
2002              buffer.pBuffer =
2003                 (gfxptr_t)(swr_resource_data(ctx->so_targets[i]->buffer) +
2004                              ctx->so_targets[i]->buffer_offset);
2005              buffer.bufferSize = ctx->so_targets[i]->buffer_size >> 2;
2006              buffer.pitch = stream_output->stride[i];
2007              buffer.streamOffset = 0;
2008 	 }
2009 
2010          ctx->api.pfnSwrSetSoBuffers(ctx->swrContext, &buffer, i);
2011       }
2012    }
2013 
2014 
2015    if (ctx->dirty & (SWR_NEW_CLIP | SWR_NEW_RASTERIZER | SWR_NEW_VS)) {
2016       // shader exporting clip distances overrides all user clip planes
2017       if (ctx->rasterizer->clip_plane_enable &&
2018           !swr_get_last_fe(ctx)->num_written_clipdistance)
2019       {
2020          swr_draw_context *pDC = &ctx->swrDC;
2021          memcpy(pDC->userClipPlanes,
2022                 ctx->clip.ucp,
2023                 sizeof(pDC->userClipPlanes));
2024       }
2025    }
2026 
2027    // set up backend state
2028    SWR_BACKEND_STATE backendState = {0};
2029    if (ctx->gs) {
2030       backendState.numAttributes = ctx->gs->info.base.num_outputs - 1;
2031    } else
2032    if (ctx->tes) {
2033       backendState.numAttributes = ctx->tes->info.base.num_outputs - 1;
2034       // no case for TCS, because if TCS is active, TES must be active
2035       // as well - pipeline stages after tessellation does not support patches
2036    }  else {
2037       backendState.numAttributes = ctx->vs->info.base.num_outputs - 1;
2038       if (ctx->fs->info.base.uses_primid) {
2039          backendState.numAttributes++;
2040          backendState.swizzleEnable = true;
2041          for (unsigned i = 0; i < sizeof(backendState.numComponents); i++) {
2042             backendState.swizzleMap[i].sourceAttrib = i;
2043          }
2044          backendState.swizzleMap[ctx->vs->info.base.num_outputs - 1].constantSource =
2045             SWR_CONSTANT_SOURCE_PRIM_ID;
2046          backendState.swizzleMap[ctx->vs->info.base.num_outputs - 1].componentOverrideMask = 1;
2047       }
2048    }
2049    if (ctx->rasterizer->sprite_coord_enable)
2050       backendState.numAttributes++;
2051 
2052    backendState.numAttributes = std::min((size_t)backendState.numAttributes,
2053                                          sizeof(backendState.numComponents));
2054    for (unsigned i = 0; i < backendState.numAttributes; i++)
2055       backendState.numComponents[i] = 4;
2056    backendState.constantInterpolationMask = ctx->fs->constantMask |
2057       (ctx->rasterizer->flatshade ? ctx->fs->flatConstantMask : 0);
2058    backendState.pointSpriteTexCoordMask = ctx->fs->pointSpriteMask;
2059 
2060    struct tgsi_shader_info *pLastFE = swr_get_last_fe(ctx);
2061 
2062    backendState.readRenderTargetArrayIndex = pLastFE->writes_layer;
2063    backendState.readViewportArrayIndex = pLastFE->writes_viewport_index;
2064    backendState.vertexAttribOffset = VERTEX_ATTRIB_START_SLOT; // TODO: optimize
2065 
2066    backendState.clipDistanceMask =
2067       pLastFE->num_written_clipdistance ?
2068       pLastFE->clipdist_writemask & ctx->rasterizer->clip_plane_enable :
2069       ctx->rasterizer->clip_plane_enable;
2070 
2071    backendState.cullDistanceMask =
2072       pLastFE->culldist_writemask << pLastFE->num_written_clipdistance;
2073 
2074    // Assume old layout of SGV, POSITION, CLIPCULL, ATTRIB
2075    backendState.vertexClipCullOffset = backendState.vertexAttribOffset - 2;
2076 
2077    ctx->api.pfnSwrSetBackendState(ctx->swrContext, &backendState);
2078 
2079    /* Ensure that any in-progress attachment change StoreTiles finish */
2080    if (swr_is_fence_pending(screen->flush_fence))
2081       swr_fence_finish(pipe->screen, NULL, screen->flush_fence, 0);
2082 
2083    /* Finally, update the in-use status of all resources involved in draw */
2084    swr_update_resource_status(pipe, p_draw_info);
2085 
2086    ctx->dirty = post_update_dirty_flags;
2087 }
2088 
2089 
2090 static struct pipe_stream_output_target *
swr_create_so_target(struct pipe_context * pipe,struct pipe_resource * buffer,unsigned buffer_offset,unsigned buffer_size)2091 swr_create_so_target(struct pipe_context *pipe,
2092                      struct pipe_resource *buffer,
2093                      unsigned buffer_offset,
2094                      unsigned buffer_size)
2095 {
2096    struct pipe_stream_output_target *target;
2097 
2098    target = CALLOC_STRUCT(pipe_stream_output_target);
2099    if (!target)
2100       return NULL;
2101 
2102    target->context = pipe;
2103    target->reference.count = 1;
2104    pipe_resource_reference(&target->buffer, buffer);
2105    target->buffer_offset = buffer_offset;
2106    target->buffer_size = buffer_size;
2107    return target;
2108 }
2109 
2110 static void
swr_destroy_so_target(struct pipe_context * pipe,struct pipe_stream_output_target * target)2111 swr_destroy_so_target(struct pipe_context *pipe,
2112                       struct pipe_stream_output_target *target)
2113 {
2114    pipe_resource_reference(&target->buffer, NULL);
2115    FREE(target);
2116 }
2117 
2118 static void
swr_set_so_targets(struct pipe_context * pipe,unsigned num_targets,struct pipe_stream_output_target ** targets,const unsigned * offsets)2119 swr_set_so_targets(struct pipe_context *pipe,
2120                    unsigned num_targets,
2121                    struct pipe_stream_output_target **targets,
2122                    const unsigned *offsets)
2123 {
2124    struct swr_context *swr = swr_context(pipe);
2125    uint32_t i;
2126 
2127    assert(num_targets <= MAX_SO_STREAMS);
2128 
2129    for (i = 0; i < num_targets; i++) {
2130       pipe_so_target_reference(
2131          (struct pipe_stream_output_target **)&swr->so_targets[i],
2132          targets[i]);
2133    }
2134 
2135    for (/* fall-through */; i < swr->num_so_targets; i++) {
2136       pipe_so_target_reference(
2137          (struct pipe_stream_output_target **)&swr->so_targets[i], NULL);
2138    }
2139 
2140    swr->num_so_targets = num_targets;
2141    swr->swrDC.soPrims = &swr->so_primCounter;
2142 
2143    swr->dirty |= SWR_NEW_SO;
2144 }
2145 
2146 
2147 void
swr_state_init(struct pipe_context * pipe)2148 swr_state_init(struct pipe_context *pipe)
2149 {
2150    pipe->create_blend_state = swr_create_blend_state;
2151    pipe->bind_blend_state = swr_bind_blend_state;
2152    pipe->delete_blend_state = swr_delete_blend_state;
2153 
2154    pipe->create_depth_stencil_alpha_state = swr_create_depth_stencil_state;
2155    pipe->bind_depth_stencil_alpha_state = swr_bind_depth_stencil_state;
2156    pipe->delete_depth_stencil_alpha_state = swr_delete_depth_stencil_state;
2157 
2158    pipe->create_rasterizer_state = swr_create_rasterizer_state;
2159    pipe->bind_rasterizer_state = swr_bind_rasterizer_state;
2160    pipe->delete_rasterizer_state = swr_delete_rasterizer_state;
2161 
2162    pipe->create_sampler_state = swr_create_sampler_state;
2163    pipe->bind_sampler_states = swr_bind_sampler_states;
2164    pipe->delete_sampler_state = swr_delete_sampler_state;
2165 
2166    pipe->create_sampler_view = swr_create_sampler_view;
2167    pipe->set_sampler_views = swr_set_sampler_views;
2168    pipe->sampler_view_destroy = swr_sampler_view_destroy;
2169 
2170    pipe->create_vs_state = swr_create_vs_state;
2171    pipe->bind_vs_state = swr_bind_vs_state;
2172    pipe->delete_vs_state = swr_delete_vs_state;
2173 
2174    pipe->create_fs_state = swr_create_fs_state;
2175    pipe->bind_fs_state = swr_bind_fs_state;
2176    pipe->delete_fs_state = swr_delete_fs_state;
2177 
2178    pipe->create_gs_state = swr_create_gs_state;
2179    pipe->bind_gs_state = swr_bind_gs_state;
2180    pipe->delete_gs_state = swr_delete_gs_state;
2181 
2182    pipe->create_tcs_state = swr_create_tcs_state;
2183    pipe->bind_tcs_state = swr_bind_tcs_state;
2184    pipe->delete_tcs_state = swr_delete_tcs_state;
2185 
2186    pipe->create_tes_state = swr_create_tes_state;
2187    pipe->bind_tes_state = swr_bind_tes_state;
2188    pipe->delete_tes_state = swr_delete_tes_state;
2189 
2190    pipe->set_constant_buffer = swr_set_constant_buffer;
2191 
2192    pipe->create_vertex_elements_state = swr_create_vertex_elements_state;
2193    pipe->bind_vertex_elements_state = swr_bind_vertex_elements_state;
2194    pipe->delete_vertex_elements_state = swr_delete_vertex_elements_state;
2195 
2196    pipe->set_vertex_buffers = swr_set_vertex_buffers;
2197 
2198    pipe->set_polygon_stipple = swr_set_polygon_stipple;
2199    pipe->set_clip_state = swr_set_clip_state;
2200    pipe->set_scissor_states = swr_set_scissor_states;
2201    pipe->set_viewport_states = swr_set_viewport_states;
2202 
2203    pipe->set_framebuffer_state = swr_set_framebuffer_state;
2204 
2205    pipe->set_blend_color = swr_set_blend_color;
2206    pipe->set_stencil_ref = swr_set_stencil_ref;
2207 
2208    pipe->set_sample_mask = swr_set_sample_mask;
2209    pipe->get_sample_position = swr_get_sample_position;
2210 
2211    pipe->create_stream_output_target = swr_create_so_target;
2212    pipe->stream_output_target_destroy = swr_destroy_so_target;
2213    pipe->set_stream_output_targets = swr_set_so_targets;
2214 }
2215