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