• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2011-2013 Luc Verhaegen <libv@skynet.be>
3  * Copyright (c) 2017-2019 Lima Project
4  *
5  * Permission is hereby granted, free of charge, to any person obtaining a
6  * copy of this software and associated documentation files (the "Software"),
7  * to deal in the Software without restriction, including without limitation
8  * the rights to use, copy, modify, merge, publish, distribute, sub license,
9  * and/or sell copies of the Software, and to permit persons to whom the
10  * Software is furnished to do so, subject to the following conditions:
11  *
12  * The above copyright notice and this permission notice (including the
13  * next paragraph) shall be included in all copies or substantial portions
14  * of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
19  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
22  * DEALINGS IN THE SOFTWARE.
23  *
24  */
25 
26 #include "util/format/u_format.h"
27 #include "util/u_debug.h"
28 #include "util/u_draw.h"
29 #include "util/half_float.h"
30 #include "util/u_helpers.h"
31 #include "util/u_inlines.h"
32 #include "util/u_pack_color.h"
33 #include "util/u_split_draw.h"
34 #include "util/u_upload_mgr.h"
35 #include "util/u_prim.h"
36 #include "util/u_vbuf.h"
37 #include "util/hash_table.h"
38 
39 #include "lima_context.h"
40 #include "lima_screen.h"
41 #include "lima_resource.h"
42 #include "lima_program.h"
43 #include "lima_bo.h"
44 #include "lima_job.h"
45 #include "lima_texture.h"
46 #include "lima_util.h"
47 #include "lima_gpu.h"
48 
49 #include "pan_minmax_cache.h"
50 
51 #include <drm-uapi/lima_drm.h>
52 
53 static void
lima_clip_scissor_to_viewport(struct lima_context * ctx)54 lima_clip_scissor_to_viewport(struct lima_context *ctx)
55 {
56    struct lima_context_framebuffer *fb = &ctx->framebuffer;
57    struct pipe_scissor_state *cscissor = &ctx->clipped_scissor;
58    int viewport_left, viewport_right, viewport_bottom, viewport_top;
59 
60    if (ctx->rasterizer && ctx->rasterizer->base.scissor) {
61       struct pipe_scissor_state *scissor = &ctx->scissor;
62       cscissor->minx = scissor->minx;
63       cscissor->maxx = scissor->maxx;
64       cscissor->miny = scissor->miny;
65       cscissor->maxy = scissor->maxy;
66    } else {
67       cscissor->minx = 0;
68       cscissor->maxx = fb->base.width;
69       cscissor->miny = 0;
70       cscissor->maxy = fb->base.height;
71    }
72 
73    viewport_left = MAX2(ctx->viewport.left, 0);
74    cscissor->minx = MAX2(cscissor->minx, viewport_left);
75    viewport_right = MIN2(MAX2(ctx->viewport.right, 0), fb->base.width);
76    cscissor->maxx = MIN2(cscissor->maxx, viewport_right);
77    if (cscissor->minx > cscissor->maxx)
78       cscissor->minx = cscissor->maxx;
79 
80    viewport_bottom = MAX2(ctx->viewport.bottom, 0);
81    cscissor->miny = MAX2(cscissor->miny, viewport_bottom);
82    viewport_top = MIN2(MAX2(ctx->viewport.top, 0), fb->base.height);
83    cscissor->maxy = MIN2(cscissor->maxy, viewport_top);
84    if (cscissor->miny > cscissor->maxy)
85       cscissor->miny = cscissor->maxy;
86 }
87 
88 static void
lima_extend_viewport(struct lima_context * ctx,const struct pipe_draw_info * info)89 lima_extend_viewport(struct lima_context *ctx, const struct pipe_draw_info *info)
90 {
91    /* restore the original values */
92    ctx->ext_viewport.left = ctx->viewport.left;
93    ctx->ext_viewport.right = ctx->viewport.right;
94    ctx->ext_viewport.bottom = ctx->viewport.bottom;
95    ctx->ext_viewport.top = ctx->viewport.top;
96 
97    if (info->mode != PIPE_PRIM_LINES)
98       return;
99 
100    if (!ctx->rasterizer)
101       return;
102 
103    float line_width = ctx->rasterizer->base.line_width;
104 
105    if (line_width == 1.0f)
106       return;
107 
108    ctx->ext_viewport.left = ctx->viewport.left - line_width / 2;
109    ctx->ext_viewport.right = ctx->viewport.right + line_width / 2;
110    ctx->ext_viewport.bottom = ctx->viewport.bottom - line_width / 2;
111    ctx->ext_viewport.top = ctx->viewport.top + line_width / 2;
112 }
113 
114 static bool
lima_is_scissor_zero(struct lima_context * ctx)115 lima_is_scissor_zero(struct lima_context *ctx)
116 {
117    struct pipe_scissor_state *cscissor = &ctx->clipped_scissor;
118 
119    return cscissor->minx == cscissor->maxx || cscissor->miny == cscissor->maxy;
120 }
121 
122 static void
lima_update_job_wb(struct lima_context * ctx,unsigned buffers)123 lima_update_job_wb(struct lima_context *ctx, unsigned buffers)
124 {
125    struct lima_job *job = lima_job_get(ctx);
126    struct lima_context_framebuffer *fb = &ctx->framebuffer;
127 
128    /* add to job when the buffer is dirty and resolve is clear (not added before) */
129    if (fb->base.nr_cbufs && (buffers & PIPE_CLEAR_COLOR0) &&
130        !(job->resolve & PIPE_CLEAR_COLOR0)) {
131       struct lima_resource *res = lima_resource(fb->base.cbufs[0]->texture);
132       lima_flush_job_accessing_bo(ctx, res->bo, true);
133       _mesa_hash_table_insert(ctx->write_jobs, &res->base, job);
134       lima_job_add_bo(job, LIMA_PIPE_PP, res->bo, LIMA_SUBMIT_BO_WRITE);
135    }
136 
137    /* add to job when the buffer is dirty and resolve is clear (not added before) */
138    if (fb->base.zsbuf && (buffers & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL)) &&
139        !(job->resolve & (PIPE_CLEAR_DEPTH | PIPE_CLEAR_STENCIL))) {
140       struct lima_resource *res = lima_resource(fb->base.zsbuf->texture);
141       lima_flush_job_accessing_bo(ctx, res->bo, true);
142       _mesa_hash_table_insert(ctx->write_jobs, &res->base, job);
143       lima_job_add_bo(job, LIMA_PIPE_PP, res->bo, LIMA_SUBMIT_BO_WRITE);
144    }
145 
146    job->resolve |= buffers;
147 }
148 
149 static void
lima_clear(struct pipe_context * pctx,unsigned buffers,const struct pipe_scissor_state * scissor_state,const union pipe_color_union * color,double depth,unsigned stencil)150 lima_clear(struct pipe_context *pctx, unsigned buffers, const struct pipe_scissor_state *scissor_state,
151            const union pipe_color_union *color, double depth, unsigned stencil)
152 {
153    struct lima_context *ctx = lima_context(pctx);
154    struct lima_job *job = lima_job_get(ctx);
155 
156    /* flush if this job already contains any draw, otherwise multi clear can be
157     * combined into a single job */
158    if (lima_job_has_draw_pending(job)) {
159       lima_do_job(job);
160       job = lima_job_get(ctx);
161    }
162 
163    lima_update_job_wb(ctx, buffers);
164 
165    /* no need to reload if cleared */
166    if (ctx->framebuffer.base.nr_cbufs && (buffers & PIPE_CLEAR_COLOR0)) {
167       struct lima_surface *surf = lima_surface(ctx->framebuffer.base.cbufs[0]);
168       surf->reload &= ~PIPE_CLEAR_COLOR0;
169    }
170 
171    struct lima_job_clear *clear = &job->clear;
172    clear->buffers = buffers;
173 
174    if (buffers & PIPE_CLEAR_COLOR0) {
175       clear->color_8pc =
176          ((uint32_t)float_to_ubyte(color->f[3]) << 24) |
177          ((uint32_t)float_to_ubyte(color->f[2]) << 16) |
178          ((uint32_t)float_to_ubyte(color->f[1]) << 8) |
179          float_to_ubyte(color->f[0]);
180 
181       clear->color_16pc =
182          ((uint64_t)float_to_ushort(color->f[3]) << 48) |
183          ((uint64_t)float_to_ushort(color->f[2]) << 32) |
184          ((uint64_t)float_to_ushort(color->f[1]) << 16) |
185          float_to_ushort(color->f[0]);
186    }
187 
188    struct lima_surface *zsbuf = lima_surface(ctx->framebuffer.base.zsbuf);
189 
190    if (buffers & PIPE_CLEAR_DEPTH) {
191       clear->depth = util_pack_z(PIPE_FORMAT_Z24X8_UNORM, depth);
192       if (zsbuf)
193          zsbuf->reload &= ~PIPE_CLEAR_DEPTH;
194    }
195 
196    if (buffers & PIPE_CLEAR_STENCIL) {
197       clear->stencil = stencil;
198       if (zsbuf)
199          zsbuf->reload &= ~PIPE_CLEAR_STENCIL;
200    }
201 
202    ctx->dirty |= LIMA_CONTEXT_DIRTY_CLEAR;
203 
204    lima_damage_rect_union(&job->damage_rect,
205                           0, ctx->framebuffer.base.width,
206                           0, ctx->framebuffer.base.height);
207 }
208 
209 enum lima_attrib_type {
210    LIMA_ATTRIB_FLOAT = 0x000,
211    LIMA_ATTRIB_I32   = 0x001,
212    LIMA_ATTRIB_U32   = 0x002,
213    LIMA_ATTRIB_FP16  = 0x003,
214    LIMA_ATTRIB_I16   = 0x004,
215    LIMA_ATTRIB_U16   = 0x005,
216    LIMA_ATTRIB_I8    = 0x006,
217    LIMA_ATTRIB_U8    = 0x007,
218    LIMA_ATTRIB_I8N   = 0x008,
219    LIMA_ATTRIB_U8N   = 0x009,
220    LIMA_ATTRIB_I16N  = 0x00A,
221    LIMA_ATTRIB_U16N  = 0x00B,
222    LIMA_ATTRIB_I32N  = 0x00D,
223    LIMA_ATTRIB_U32N  = 0x00E,
224    LIMA_ATTRIB_FIXED = 0x101
225 };
226 
227 static enum lima_attrib_type
lima_pipe_format_to_attrib_type(enum pipe_format format)228 lima_pipe_format_to_attrib_type(enum pipe_format format)
229 {
230    const struct util_format_description *desc = util_format_description(format);
231    int i = util_format_get_first_non_void_channel(format);
232    const struct util_format_channel_description *c = desc->channel + i;
233 
234    switch (c->type) {
235    case UTIL_FORMAT_TYPE_FLOAT:
236       if (c->size == 16)
237          return LIMA_ATTRIB_FP16;
238       else
239          return LIMA_ATTRIB_FLOAT;
240    case UTIL_FORMAT_TYPE_FIXED:
241       return LIMA_ATTRIB_FIXED;
242    case UTIL_FORMAT_TYPE_SIGNED:
243       if (c->size == 8) {
244          if (c->normalized)
245             return LIMA_ATTRIB_I8N;
246          else
247             return LIMA_ATTRIB_I8;
248       }
249       else if (c->size == 16) {
250          if (c->normalized)
251             return LIMA_ATTRIB_I16N;
252          else
253             return LIMA_ATTRIB_I16;
254       }
255       else if (c->size == 32) {
256          if (c->normalized)
257             return LIMA_ATTRIB_I32N;
258          else
259             return LIMA_ATTRIB_I32;
260       }
261       break;
262    case UTIL_FORMAT_TYPE_UNSIGNED:
263       if (c->size == 8) {
264          if (c->normalized)
265             return LIMA_ATTRIB_U8N;
266          else
267             return LIMA_ATTRIB_U8;
268       }
269       else if (c->size == 16) {
270          if (c->normalized)
271             return LIMA_ATTRIB_U16N;
272          else
273             return LIMA_ATTRIB_U16;
274       }
275       else if (c->size == 32) {
276          if (c->normalized)
277             return LIMA_ATTRIB_U32N;
278          else
279             return LIMA_ATTRIB_U32;
280       }
281       break;
282    }
283 
284    return LIMA_ATTRIB_FLOAT;
285 }
286 
287 static void
lima_pack_vs_cmd(struct lima_context * ctx,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw)288 lima_pack_vs_cmd(struct lima_context *ctx, const struct pipe_draw_info *info,
289                  const struct pipe_draw_start_count_bias *draw)
290 {
291    struct lima_context_constant_buffer *ccb =
292       ctx->const_buffer + PIPE_SHADER_VERTEX;
293    struct lima_vs_compiled_shader *vs = ctx->vs;
294    struct lima_job *job = lima_job_get(ctx);
295 
296    VS_CMD_BEGIN(&job->vs_cmd_array, 24);
297 
298    if (!info->index_size) {
299       VS_CMD_ARRAYS_SEMAPHORE_BEGIN_1();
300       VS_CMD_ARRAYS_SEMAPHORE_BEGIN_2();
301    }
302    int uniform_size = MIN2(vs->state.uniform_size, ccb->size);
303 
304    int size = uniform_size + vs->state.constant_size + 32;
305    VS_CMD_UNIFORMS_ADDRESS(
306       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_uniform),
307       align(size, 16));
308 
309    VS_CMD_SHADER_ADDRESS(ctx->vs->bo->va, ctx->vs->state.shader_size);
310    VS_CMD_SHADER_INFO(ctx->vs->state.prefetch, ctx->vs->state.shader_size);
311 
312    int num_outputs = ctx->vs->state.num_outputs;
313    int num_attributes = ctx->vertex_elements->num_elements;
314    VS_CMD_VARYING_ATTRIBUTE_COUNT(num_outputs, MAX2(1, num_attributes));
315 
316    VS_CMD_UNKNOWN1();
317 
318    VS_CMD_ATTRIBUTES_ADDRESS(
319       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_attribute_info),
320       MAX2(1, num_attributes));
321 
322    VS_CMD_VARYINGS_ADDRESS(
323       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_varying_info),
324       num_outputs);
325 
326    unsigned num = info->index_size ? (ctx->max_index - ctx->min_index + 1) : draw->count;
327    VS_CMD_DRAW(num, info->index_size);
328 
329    VS_CMD_UNKNOWN2();
330 
331    VS_CMD_ARRAYS_SEMAPHORE_END(info->index_size);
332 
333    VS_CMD_END();
334 }
335 
336 static void
lima_pack_plbu_cmd(struct lima_context * ctx,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw)337 lima_pack_plbu_cmd(struct lima_context *ctx, const struct pipe_draw_info *info,
338                    const struct pipe_draw_start_count_bias *draw)
339 {
340    struct lima_vs_compiled_shader *vs = ctx->vs;
341    struct pipe_scissor_state *cscissor = &ctx->clipped_scissor;
342    struct lima_job *job = lima_job_get(ctx);
343    PLBU_CMD_BEGIN(&job->plbu_cmd_array, 32);
344 
345    PLBU_CMD_VIEWPORT_LEFT(fui(ctx->ext_viewport.left));
346    PLBU_CMD_VIEWPORT_RIGHT(fui(ctx->ext_viewport.right));
347    PLBU_CMD_VIEWPORT_BOTTOM(fui(ctx->ext_viewport.bottom));
348    PLBU_CMD_VIEWPORT_TOP(fui(ctx->ext_viewport.top));
349 
350    if (!info->index_size)
351       PLBU_CMD_ARRAYS_SEMAPHORE_BEGIN();
352 
353    int cf = ctx->rasterizer->base.cull_face;
354    int ccw = ctx->rasterizer->base.front_ccw;
355    uint32_t cull = 0;
356    bool force_point_size = false;
357 
358    if (cf != PIPE_FACE_NONE) {
359       if (cf & PIPE_FACE_FRONT)
360          cull |= ccw ? 0x00040000 : 0x00020000;
361       if (cf & PIPE_FACE_BACK)
362          cull |= ccw ? 0x00020000 : 0x00040000;
363    }
364 
365    /* Specify point size with PLBU command if shader doesn't write */
366    if (info->mode == PIPE_PRIM_POINTS && ctx->vs->state.point_size_idx == -1)
367       force_point_size = true;
368 
369    /* Specify line width with PLBU command for lines */
370    if (info->mode > PIPE_PRIM_POINTS && info->mode < PIPE_PRIM_TRIANGLES)
371       force_point_size = true;
372 
373    PLBU_CMD_PRIMITIVE_SETUP(force_point_size, cull, info->index_size);
374 
375    PLBU_CMD_RSW_VERTEX_ARRAY(
376       lima_ctx_buff_va(ctx, lima_ctx_buff_pp_plb_rsw),
377       ctx->gp_output->va);
378 
379    /* TODO
380     * - we should set it only for the first draw that enabled the scissor and for
381     *   latter draw only if scissor is dirty
382     */
383 
384    assert(cscissor->minx < cscissor->maxx && cscissor->miny < cscissor->maxy);
385    PLBU_CMD_SCISSORS(cscissor->minx, cscissor->maxx, cscissor->miny, cscissor->maxy);
386 
387    lima_damage_rect_union(&job->damage_rect, cscissor->minx, cscissor->maxx,
388                           cscissor->miny, cscissor->maxy);
389 
390    PLBU_CMD_UNKNOWN1();
391 
392    PLBU_CMD_DEPTH_RANGE_NEAR(fui(ctx->viewport.near));
393    PLBU_CMD_DEPTH_RANGE_FAR(fui(ctx->viewport.far));
394 
395    if ((info->mode == PIPE_PRIM_POINTS && ctx->vs->state.point_size_idx == -1) ||
396        ((info->mode >= PIPE_PRIM_LINES) && (info->mode < PIPE_PRIM_TRIANGLES)))
397    {
398       uint32_t v = info->mode == PIPE_PRIM_POINTS ?
399          fui(ctx->rasterizer->base.point_size) : fui(ctx->rasterizer->base.line_width);
400       PLBU_CMD_LOW_PRIM_SIZE(v);
401    }
402 
403    if (info->index_size) {
404       PLBU_CMD_INDEXED_DEST(ctx->gp_output->va);
405       if (vs->state.point_size_idx != -1)
406          PLBU_CMD_INDEXED_PT_SIZE(ctx->gp_output->va + ctx->gp_output_point_size_offt);
407 
408       PLBU_CMD_INDICES(ctx->index_res->bo->va + draw->start * info->index_size + ctx->index_offset);
409    }
410    else {
411       /* can this make the attribute info static? */
412       PLBU_CMD_DRAW_ARRAYS(info->mode, draw->start, draw->count);
413    }
414 
415    PLBU_CMD_ARRAYS_SEMAPHORE_END();
416 
417    if (info->index_size)
418       PLBU_CMD_DRAW_ELEMENTS(info->mode, ctx->min_index, draw->count);
419 
420    PLBU_CMD_END();
421 }
422 
423 static int
lima_blend_func(enum pipe_blend_func pipe)424 lima_blend_func(enum pipe_blend_func pipe)
425 {
426    switch (pipe) {
427    case PIPE_BLEND_ADD:
428       return 2;
429    case PIPE_BLEND_SUBTRACT:
430       return 0;
431    case PIPE_BLEND_REVERSE_SUBTRACT:
432       return 1;
433    case PIPE_BLEND_MIN:
434       return 4;
435    case PIPE_BLEND_MAX:
436       return 5;
437    }
438    return -1;
439 }
440 
441 static int
lima_blend_factor(enum pipe_blendfactor pipe)442 lima_blend_factor(enum pipe_blendfactor pipe)
443 {
444    /* Bits 0-2 indicate the blendfactor type,
445     * Bit 3 is set if blendfactor is inverted
446     * Bit 4 is set if blendfactor has alpha */
447    switch (pipe) {
448    case PIPE_BLENDFACTOR_SRC_COLOR:
449       return 0 << 4 | 0 << 3 | 0;
450    case PIPE_BLENDFACTOR_SRC_ALPHA:
451       return 1 << 4 | 0 << 3 | 0;
452    case PIPE_BLENDFACTOR_INV_SRC_COLOR:
453       return 0 << 4 | 1 << 3 | 0;
454    case PIPE_BLENDFACTOR_INV_SRC_ALPHA:
455       return 1 << 4 | 1 << 3 | 0;
456 
457    case PIPE_BLENDFACTOR_DST_COLOR:
458       return 0 << 4 | 0 << 3 | 1;
459    case PIPE_BLENDFACTOR_DST_ALPHA:
460       return 1 << 4 | 0 << 3 | 1;
461    case PIPE_BLENDFACTOR_INV_DST_COLOR:
462       return 0 << 4 | 1 << 3 | 1;
463    case PIPE_BLENDFACTOR_INV_DST_ALPHA:
464       return 1 << 4 | 1 << 3 | 1;
465 
466    case PIPE_BLENDFACTOR_CONST_COLOR:
467       return 0 << 4 | 0 << 3 | 2;
468    case PIPE_BLENDFACTOR_CONST_ALPHA:
469       return 1 << 4 | 0 << 3 | 2;
470    case PIPE_BLENDFACTOR_INV_CONST_COLOR:
471       return 0 << 4 | 1 << 3 | 2;
472    case PIPE_BLENDFACTOR_INV_CONST_ALPHA:
473       return 1 << 4 | 1 << 3 | 2;
474 
475    case PIPE_BLENDFACTOR_ZERO:
476       return 0 << 4 | 0 << 3 | 3;
477    case PIPE_BLENDFACTOR_ONE:
478       return 0 << 4 | 1 << 3 | 3;
479 
480    case PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE:
481       return 0 << 4 | 0 << 3 | 4;
482 
483    case PIPE_BLENDFACTOR_SRC1_COLOR:
484       return 0 << 4 | 0 << 3 | 5;
485    case PIPE_BLENDFACTOR_SRC1_ALPHA:
486       return 1 << 4 | 0 << 3 | 5;
487    case PIPE_BLENDFACTOR_INV_SRC1_COLOR:
488       return 0 << 4 | 1 << 3 | 5;
489    case PIPE_BLENDFACTOR_INV_SRC1_ALPHA:
490       return 1 << 4 | 1 << 3 | 5;
491    }
492    return -1;
493 }
494 
495 static int
lima_calculate_alpha_blend(enum pipe_blend_func rgb_func,enum pipe_blend_func alpha_func,enum pipe_blendfactor rgb_src_factor,enum pipe_blendfactor rgb_dst_factor,enum pipe_blendfactor alpha_src_factor,enum pipe_blendfactor alpha_dst_factor)496 lima_calculate_alpha_blend(enum pipe_blend_func rgb_func, enum pipe_blend_func alpha_func,
497                            enum pipe_blendfactor rgb_src_factor, enum pipe_blendfactor rgb_dst_factor,
498                            enum pipe_blendfactor alpha_src_factor, enum pipe_blendfactor alpha_dst_factor)
499 {
500    /* PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE has to be changed to PIPE_BLENDFACTOR_ONE
501     * if it is set for alpha_src or alpha_dst.
502     */
503    if (alpha_src_factor == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE)
504       alpha_src_factor = PIPE_BLENDFACTOR_ONE;
505 
506    if (alpha_dst_factor == PIPE_BLENDFACTOR_SRC_ALPHA_SATURATE)
507       alpha_dst_factor = PIPE_BLENDFACTOR_ONE;
508 
509    /* MIN and MAX ops actually do OP(As * S + Ad * D, Ad), so
510     * we need to set S to 1 and D to 0 to get correct result */
511    if (alpha_func == PIPE_BLEND_MIN ||
512        alpha_func == PIPE_BLEND_MAX) {
513       alpha_src_factor = PIPE_BLENDFACTOR_ONE;
514       alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
515    }
516 
517    /* MIN and MAX ops actually do OP(Cs * S + Cd * D, Cd), so
518     * we need to set S to 1 and D to 0 to get correct result */
519    if (rgb_func == PIPE_BLEND_MIN ||
520        rgb_func == PIPE_BLEND_MAX) {
521       rgb_src_factor = PIPE_BLENDFACTOR_ONE;
522       rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
523    }
524 
525    return lima_blend_func(rgb_func) |
526       (lima_blend_func(alpha_func) << 3) |
527       (lima_blend_factor(rgb_src_factor) << 6) |
528       (lima_blend_factor(rgb_dst_factor) << 11) |
529       /* alpha_src and alpha_dst are 4 bit, so need to mask 5th bit */
530       ((lima_blend_factor(alpha_src_factor) & 0xf) << 16) |
531       ((lima_blend_factor(alpha_dst_factor) & 0xf) << 20) |
532       0x0C000000; /* need to check if this is GLESv1 glAlphaFunc */
533 }
534 
535 static int
lima_stencil_op(enum pipe_stencil_op pipe)536 lima_stencil_op(enum pipe_stencil_op pipe)
537 {
538    switch (pipe) {
539    case PIPE_STENCIL_OP_KEEP:
540       return 0;
541    case PIPE_STENCIL_OP_ZERO:
542       return 2;
543    case PIPE_STENCIL_OP_REPLACE:
544       return 1;
545    case PIPE_STENCIL_OP_INCR:
546       return 6;
547    case PIPE_STENCIL_OP_DECR:
548       return 7;
549    case PIPE_STENCIL_OP_INCR_WRAP:
550       return 4;
551    case PIPE_STENCIL_OP_DECR_WRAP:
552       return 5;
553    case PIPE_STENCIL_OP_INVERT:
554       return 3;
555    }
556    return -1;
557 }
558 
559 static unsigned
lima_calculate_depth_test(struct pipe_depth_stencil_alpha_state * depth,struct pipe_rasterizer_state * rst)560 lima_calculate_depth_test(struct pipe_depth_stencil_alpha_state *depth,
561                           struct pipe_rasterizer_state *rst)
562 {
563    int offset_scale = 0, offset_units = 0;
564    enum pipe_compare_func func = (depth->depth_enabled ? depth->depth_func : PIPE_FUNC_ALWAYS);
565 
566    offset_scale = CLAMP(rst->offset_scale * 4, -128, 127);
567    if (offset_scale < 0)
568       offset_scale += 0x100;
569 
570    offset_units = CLAMP(rst->offset_units * 2, -128, 127);
571    if (offset_units < 0)
572       offset_units += 0x100;
573 
574    return (depth->depth_enabled && depth->depth_writemask) |
575       ((int)func << 1) |
576       (offset_scale << 16) |
577       (offset_units << 24);
578 }
579 
580 static void
lima_pack_render_state(struct lima_context * ctx,const struct pipe_draw_info * info)581 lima_pack_render_state(struct lima_context *ctx, const struct pipe_draw_info *info)
582 {
583    struct lima_fs_compiled_shader *fs = ctx->fs;
584    struct lima_render_state *render =
585       lima_ctx_buff_alloc(ctx, lima_ctx_buff_pp_plb_rsw,
586                           sizeof(*render));
587    bool early_z = true;
588    bool pixel_kill = true;
589 
590    /* do hw support RGBA independ blend?
591     * PIPE_CAP_INDEP_BLEND_ENABLE
592     *
593     * how to handle the no cbuf only zbuf case?
594     */
595    struct pipe_rt_blend_state *rt = ctx->blend->base.rt;
596    render->blend_color_bg = float_to_ubyte(ctx->blend_color.color[2]) |
597       (float_to_ubyte(ctx->blend_color.color[1]) << 16);
598    render->blend_color_ra = float_to_ubyte(ctx->blend_color.color[0]) |
599       (float_to_ubyte(ctx->blend_color.color[3]) << 16);
600 
601    if (rt->blend_enable) {
602       render->alpha_blend = lima_calculate_alpha_blend(rt->rgb_func, rt->alpha_func,
603          rt->rgb_src_factor, rt->rgb_dst_factor,
604          rt->alpha_src_factor, rt->alpha_dst_factor);
605    }
606    else {
607       /*
608        * Special handling for blending disabled.
609        * Binary driver is generating the same alpha_value,
610        * as when we would just enable blending, without changing/setting any blend equation/params.
611        * Normaly in this case mesa would set all rt fields (func/factor) to zero.
612        */
613       render->alpha_blend = lima_calculate_alpha_blend(PIPE_BLEND_ADD, PIPE_BLEND_ADD,
614          PIPE_BLENDFACTOR_ONE, PIPE_BLENDFACTOR_ZERO,
615          PIPE_BLENDFACTOR_ONE, PIPE_BLENDFACTOR_ZERO);
616    }
617 
618    render->alpha_blend |= (rt->colormask & PIPE_MASK_RGBA) << 28;
619 
620    struct pipe_rasterizer_state *rst = &ctx->rasterizer->base;
621    render->depth_test = lima_calculate_depth_test(&ctx->zsa->base, rst);
622 
623    if (!rst->depth_clip_near || ctx->viewport.near == 0.0f)
624       render->depth_test |= 0x10; /* don't clip depth near */
625    if (!rst->depth_clip_far || ctx->viewport.far == 1.0f)
626       render->depth_test |= 0x20; /* don't clip depth far */
627 
628    if (fs->state.frag_depth_reg != -1) {
629       render->depth_test |= (fs->state.frag_depth_reg << 6);
630       /* Shader writes depth */
631       render->depth_test |= 0x801;
632    }
633 
634    ushort far, near;
635 
636    near = float_to_ushort(ctx->viewport.near);
637    far = float_to_ushort(ctx->viewport.far);
638 
639    /* overlap with plbu? any place can remove one? */
640    render->depth_range = near | (far << 16);
641 
642    struct pipe_stencil_state *stencil = ctx->zsa->base.stencil;
643    struct pipe_stencil_ref *ref = &ctx->stencil_ref;
644 
645    if (stencil[0].enabled) { /* stencil is enabled */
646       render->stencil_front = stencil[0].func |
647          (lima_stencil_op(stencil[0].fail_op) << 3) |
648          (lima_stencil_op(stencil[0].zfail_op) << 6) |
649          (lima_stencil_op(stencil[0].zpass_op) << 9) |
650          (ref->ref_value[0] << 16) |
651          (stencil[0].valuemask << 24);
652       render->stencil_back = render->stencil_front;
653       render->stencil_test = (stencil[0].writemask & 0xff) | (stencil[0].writemask & 0xff) << 8;
654       if (stencil[1].enabled) { /* two-side is enabled */
655          render->stencil_back = stencil[1].func |
656             (lima_stencil_op(stencil[1].fail_op) << 3) |
657             (lima_stencil_op(stencil[1].zfail_op) << 6) |
658             (lima_stencil_op(stencil[1].zpass_op) << 9) |
659             (ref->ref_value[1] << 16) |
660             (stencil[1].valuemask << 24);
661          render->stencil_test = (stencil[0].writemask & 0xff) | (stencil[1].writemask & 0xff) << 8;
662       }
663       /* TODO: Find out, what (render->stecil_test & 0xff000000) is */
664    }
665    else {
666       /* Default values, when stencil is disabled:
667        * stencil[0|1].valuemask = 0xff
668        * stencil[0|1].func = PIPE_FUNC_ALWAYS
669        * stencil[0|1].writemask = 0xff
670        */
671       render->stencil_front = 0xff000007;
672       render->stencil_back = 0xff000007;
673       render->stencil_test = 0x0000ffff;
674    }
675 
676    /* need more investigation */
677    if (info->mode == PIPE_PRIM_POINTS)
678       render->multi_sample = 0x00000000;
679    else if (info->mode < PIPE_PRIM_TRIANGLES)
680       render->multi_sample = 0x00000400;
681    else
682       render->multi_sample = 0x00000800;
683    if (ctx->framebuffer.base.samples)
684       render->multi_sample |= 0x68;
685    if (ctx->blend->base.alpha_to_coverage)
686       render->multi_sample |= (1 << 7);
687    if (ctx->blend->base.alpha_to_one)
688       render->multi_sample |= (1 << 8);
689    render->multi_sample |= (ctx->sample_mask << 12);
690 
691    /* Set gl_FragColor register, need to specify it 4 times */
692    render->multi_sample |= (fs->state.frag_color0_reg << 28) |
693                            (fs->state.frag_color0_reg << 24) |
694                            (fs->state.frag_color0_reg << 20) |
695                            (fs->state.frag_color0_reg << 16);
696 
697    /* alpha test */
698    if (ctx->zsa->base.alpha_enabled) {
699       render->multi_sample |= ctx->zsa->base.alpha_func;
700       render->stencil_test |= float_to_ubyte(ctx->zsa->base.alpha_ref_value) << 16;
701    } else {
702       /* func = PIPE_FUNC_ALWAYS */
703       render->multi_sample |= 0x7;
704    }
705 
706    render->shader_address =
707       ctx->fs->bo->va | (((uint32_t *)ctx->fs->bo->map)[0] & 0x1F);
708 
709    /* seems not needed */
710    render->uniforms_address = 0x00000000;
711 
712    render->textures_address = 0x00000000;
713 
714    render->aux0 = (ctx->vs->state.varying_stride >> 3);
715    render->aux1 = 0x00000000;
716    if (ctx->rasterizer->base.front_ccw)
717       render->aux1 = 0x00001000;
718 
719    if (ctx->blend->base.dither)
720       render->aux1 |= 0x00002000;
721 
722    if (fs->state.uses_discard ||
723        ctx->zsa->base.alpha_enabled ||
724        fs->state.frag_depth_reg != -1 ||
725        ctx->blend->base.alpha_to_coverage) {
726       early_z = false;
727       pixel_kill = false;
728    }
729 
730    if (rt->blend_enable)
731       pixel_kill = false;
732 
733    if ((rt->colormask & PIPE_MASK_RGBA) != PIPE_MASK_RGBA)
734       pixel_kill = false;
735 
736    if (early_z)
737       render->aux0 |= 0x300;
738 
739    if (pixel_kill)
740       render->aux0 |= 0x1000;
741 
742    if (ctx->tex_stateobj.num_samplers) {
743       render->textures_address =
744          lima_ctx_buff_va(ctx, lima_ctx_buff_pp_tex_desc);
745       render->aux0 |= ctx->tex_stateobj.num_samplers << 14;
746       render->aux0 |= 0x20;
747    }
748 
749    if (ctx->const_buffer[PIPE_SHADER_FRAGMENT].buffer) {
750       render->uniforms_address =
751          lima_ctx_buff_va(ctx, lima_ctx_buff_pp_uniform_array);
752       uint32_t size = ctx->buffer_state[lima_ctx_buff_pp_uniform].size;
753       uint32_t bits = 0;
754       if (size >= 8) {
755          bits = util_last_bit(size >> 3) - 1;
756          bits += size & u_bit_consecutive(0, bits + 3) ? 1 : 0;
757       }
758       render->uniforms_address |= bits > 0xf ? 0xf : bits;
759 
760       render->aux0 |= 0x80;
761       render->aux1 |= 0x10000;
762    }
763 
764    /* Set secondary output color */
765    if (fs->state.frag_color1_reg != -1)
766       render->aux0 |= (fs->state.frag_color1_reg << 28);
767 
768    if (ctx->vs->state.num_varyings) {
769       render->varying_types = 0x00000000;
770       render->varyings_address = ctx->gp_output->va +
771                                  ctx->gp_output_varyings_offt;
772       for (int i = 0, index = 0; i < ctx->vs->state.num_outputs; i++) {
773          int val;
774 
775          if (i == ctx->vs->state.gl_pos_idx ||
776              i == ctx->vs->state.point_size_idx)
777             continue;
778 
779          struct lima_varying_info *v = ctx->vs->state.varying + i;
780          if (v->component_size == 4)
781             val = v->components > 2 ? 0 : 1;
782          else
783             val = v->components > 2 ? 2 : 3;
784 
785          if (index < 10)
786             render->varying_types |= val << (3 * index);
787          else if (index == 10) {
788             render->varying_types |= val << 30;
789             render->varyings_address |= val >> 2;
790          }
791          else if (index == 11)
792             render->varyings_address |= val << 1;
793 
794          index++;
795       }
796    }
797    else {
798       render->varying_types = 0x00000000;
799       render->varyings_address = 0x00000000;
800    }
801 
802    struct lima_job *job = lima_job_get(ctx);
803 
804    lima_dump_command_stream_print(
805       job->dump, render, sizeof(*render),
806       false, "add render state at va %x\n",
807       lima_ctx_buff_va(ctx, lima_ctx_buff_pp_plb_rsw));
808 
809    lima_dump_rsw_command_stream_print(
810       job->dump, render, sizeof(*render),
811       lima_ctx_buff_va(ctx, lima_ctx_buff_pp_plb_rsw));
812 }
813 
814 static void
lima_update_gp_attribute_info(struct lima_context * ctx,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw)815 lima_update_gp_attribute_info(struct lima_context *ctx, const struct pipe_draw_info *info,
816                               const struct pipe_draw_start_count_bias *draw)
817 {
818    struct lima_job *job = lima_job_get(ctx);
819    struct lima_vertex_element_state *ve = ctx->vertex_elements;
820    struct lima_context_vertex_buffer *vb = &ctx->vertex_buffers;
821 
822    uint32_t *attribute =
823       lima_ctx_buff_alloc(ctx, lima_ctx_buff_gp_attribute_info,
824                           MAX2(1, ve->num_elements) * 8);
825 
826    int n = 0;
827    for (int i = 0; i < ve->num_elements; i++) {
828       struct pipe_vertex_element *pve = ve->pipe + i;
829 
830       assert(pve->vertex_buffer_index < vb->count);
831       assert(vb->enabled_mask & (1 << pve->vertex_buffer_index));
832 
833       struct pipe_vertex_buffer *pvb = vb->vb + pve->vertex_buffer_index;
834       struct lima_resource *res = lima_resource(pvb->buffer.resource);
835 
836       lima_job_add_bo(job, LIMA_PIPE_GP, res->bo, LIMA_SUBMIT_BO_READ);
837 
838       unsigned start = info->index_size ? (ctx->min_index + draw->index_bias) : draw->start;
839       attribute[n++] = res->bo->va + pvb->buffer_offset + pve->src_offset
840          + start * pvb->stride;
841       attribute[n++] = (pvb->stride << 11) |
842          (lima_pipe_format_to_attrib_type(pve->src_format) << 2) |
843          (util_format_get_nr_components(pve->src_format) - 1);
844    }
845 
846    lima_dump_command_stream_print(
847       job->dump, attribute, n * 4, false, "update attribute info at va %x\n",
848       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_attribute_info));
849 }
850 
851 static void
lima_update_gp_uniform(struct lima_context * ctx)852 lima_update_gp_uniform(struct lima_context *ctx)
853 {
854    struct lima_context_constant_buffer *ccb =
855       ctx->const_buffer + PIPE_SHADER_VERTEX;
856    struct lima_vs_compiled_shader *vs = ctx->vs;
857    int uniform_size = MIN2(vs->state.uniform_size, ccb->size);
858 
859    int size = uniform_size + vs->state.constant_size + 32;
860    void *vs_const_buff =
861       lima_ctx_buff_alloc(ctx, lima_ctx_buff_gp_uniform, size);
862 
863    if (ccb->buffer)
864       memcpy(vs_const_buff, ccb->buffer, uniform_size);
865 
866    memcpy(vs_const_buff + uniform_size,
867           ctx->viewport.transform.scale,
868           sizeof(ctx->viewport.transform.scale));
869    memcpy(vs_const_buff + uniform_size + 16,
870           ctx->viewport.transform.translate,
871           sizeof(ctx->viewport.transform.translate));
872 
873    if (vs->constant)
874       memcpy(vs_const_buff + uniform_size + 32,
875              vs->constant, vs->state.constant_size);
876 
877    struct lima_job *job = lima_job_get(ctx);
878 
879    if (lima_debug & LIMA_DEBUG_GP) {
880       float *vs_const_buff_f = vs_const_buff;
881       printf("gp uniforms:\n");
882       for (int i = 0; i < (size / sizeof(float)); i++) {
883          if ((i % 4) == 0)
884             printf("%4d:", i / 4);
885          printf(" %8.4f", vs_const_buff_f[i]);
886          if ((i % 4) == 3)
887             printf("\n");
888       }
889       printf("\n");
890    }
891 
892    lima_dump_command_stream_print(
893       job->dump, vs_const_buff, size, true,
894       "update gp uniform at va %x\n",
895       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_uniform));
896 }
897 
898 static void
lima_update_pp_uniform(struct lima_context * ctx)899 lima_update_pp_uniform(struct lima_context *ctx)
900 {
901    const float *const_buff = ctx->const_buffer[PIPE_SHADER_FRAGMENT].buffer;
902    size_t const_buff_size = ctx->const_buffer[PIPE_SHADER_FRAGMENT].size / sizeof(float);
903 
904    if (!const_buff)
905       return;
906 
907    uint16_t *fp16_const_buff =
908       lima_ctx_buff_alloc(ctx, lima_ctx_buff_pp_uniform,
909                           const_buff_size * sizeof(uint16_t));
910 
911    uint32_t *array =
912       lima_ctx_buff_alloc(ctx, lima_ctx_buff_pp_uniform_array, 4);
913 
914    for (int i = 0; i < const_buff_size; i++)
915        fp16_const_buff[i] = _mesa_float_to_half(const_buff[i]);
916 
917    *array = lima_ctx_buff_va(ctx, lima_ctx_buff_pp_uniform);
918 
919    struct lima_job *job = lima_job_get(ctx);
920 
921    lima_dump_command_stream_print(
922       job->dump, fp16_const_buff, const_buff_size * 2,
923       false, "add pp uniform data at va %x\n",
924       lima_ctx_buff_va(ctx, lima_ctx_buff_pp_uniform));
925    lima_dump_command_stream_print(
926       job->dump, array, 4, false, "add pp uniform info at va %x\n",
927       lima_ctx_buff_va(ctx, lima_ctx_buff_pp_uniform_array));
928 }
929 
930 static void
lima_update_varying(struct lima_context * ctx,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw)931 lima_update_varying(struct lima_context *ctx, const struct pipe_draw_info *info,
932                     const struct pipe_draw_start_count_bias *draw)
933 {
934    struct lima_job *job = lima_job_get(ctx);
935    struct lima_screen *screen = lima_screen(ctx->base.screen);
936    struct lima_vs_compiled_shader *vs = ctx->vs;
937    uint32_t gp_output_size;
938    unsigned num = info->index_size ? (ctx->max_index - ctx->min_index + 1) : draw->count;
939 
940    uint32_t *varying =
941       lima_ctx_buff_alloc(ctx, lima_ctx_buff_gp_varying_info,
942                           vs->state.num_outputs * 8);
943    int n = 0;
944 
945    int offset = 0;
946 
947    for (int i = 0; i < vs->state.num_outputs; i++) {
948       struct lima_varying_info *v = vs->state.varying + i;
949 
950       if (i == vs->state.gl_pos_idx ||
951           i == vs->state.point_size_idx)
952          continue;
953 
954       int size = v->component_size * 4;
955 
956       /* does component_size == 2 need to be 16 aligned? */
957       if (v->component_size == 4)
958          offset = align(offset, 16);
959 
960       v->offset = offset;
961       offset += size;
962    }
963 
964    vs->state.varying_stride = align(offset, 16);
965 
966    /* gl_Position is always present, allocate space for it */
967    gp_output_size = align(4 * 4 * num, 0x40);
968 
969    /* Allocate space for varyings if there're any */
970    if (vs->state.num_varyings) {
971       ctx->gp_output_varyings_offt = gp_output_size;
972       gp_output_size += align(vs->state.varying_stride * num, 0x40);
973    }
974 
975    /* Allocate space for gl_PointSize if it's there */
976    if (vs->state.point_size_idx != -1) {
977       ctx->gp_output_point_size_offt = gp_output_size;
978       gp_output_size += 4 * num;
979    }
980 
981    /* gp_output can be too large for the suballocator, so create a
982     * separate bo for it. The bo cache should prevent performance hit.
983     */
984    ctx->gp_output = lima_bo_create(screen, gp_output_size, 0);
985    assert(ctx->gp_output);
986    lima_job_add_bo(job, LIMA_PIPE_GP, ctx->gp_output, LIMA_SUBMIT_BO_WRITE);
987    lima_job_add_bo(job, LIMA_PIPE_PP, ctx->gp_output, LIMA_SUBMIT_BO_READ);
988 
989    for (int i = 0; i < vs->state.num_outputs; i++) {
990       struct lima_varying_info *v = vs->state.varying + i;
991 
992       if (i == vs->state.gl_pos_idx) {
993          /* gl_Position */
994          varying[n++] = ctx->gp_output->va;
995          varying[n++] = 0x8020;
996       } else if (i == vs->state.point_size_idx) {
997          /* gl_PointSize */
998          varying[n++] = ctx->gp_output->va + ctx->gp_output_point_size_offt;
999          varying[n++] = 0x2021;
1000       } else {
1001          /* Varying */
1002          varying[n++] = ctx->gp_output->va + ctx->gp_output_varyings_offt +
1003                         v->offset;
1004          varying[n++] = (vs->state.varying_stride << 11) | (v->components - 1) |
1005             (v->component_size == 2 ? 0x0C : 0);
1006       }
1007    }
1008 
1009    lima_dump_command_stream_print(
1010       job->dump, varying, n * 4, false, "update varying info at va %x\n",
1011       lima_ctx_buff_va(ctx, lima_ctx_buff_gp_varying_info));
1012 }
1013 
1014 static void
lima_draw_vbo_update(struct pipe_context * pctx,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw)1015 lima_draw_vbo_update(struct pipe_context *pctx,
1016                      const struct pipe_draw_info *info,
1017                      const struct pipe_draw_start_count_bias *draw)
1018 {
1019    struct lima_context *ctx = lima_context(pctx);
1020    struct lima_context_framebuffer *fb = &ctx->framebuffer;
1021    unsigned buffers = 0;
1022 
1023    if (fb->base.zsbuf) {
1024       if (ctx->zsa->base.depth_enabled)
1025          buffers |= PIPE_CLEAR_DEPTH;
1026       if (ctx->zsa->base.stencil[0].enabled ||
1027           ctx->zsa->base.stencil[1].enabled)
1028          buffers |= PIPE_CLEAR_STENCIL;
1029    }
1030 
1031    if (fb->base.nr_cbufs)
1032       buffers |= PIPE_CLEAR_COLOR0;
1033 
1034    lima_update_job_wb(ctx, buffers);
1035 
1036    lima_update_gp_attribute_info(ctx, info, draw);
1037 
1038    if ((ctx->dirty & LIMA_CONTEXT_DIRTY_CONST_BUFF &&
1039         ctx->const_buffer[PIPE_SHADER_VERTEX].dirty) ||
1040        ctx->dirty & LIMA_CONTEXT_DIRTY_VIEWPORT ||
1041        ctx->dirty & LIMA_CONTEXT_DIRTY_COMPILED_VS) {
1042       lima_update_gp_uniform(ctx);
1043       ctx->const_buffer[PIPE_SHADER_VERTEX].dirty = false;
1044    }
1045 
1046    lima_update_varying(ctx, info, draw);
1047 
1048    lima_pack_vs_cmd(ctx, info, draw);
1049 
1050    if (ctx->dirty & LIMA_CONTEXT_DIRTY_CONST_BUFF &&
1051        ctx->const_buffer[PIPE_SHADER_FRAGMENT].dirty) {
1052       lima_update_pp_uniform(ctx);
1053       ctx->const_buffer[PIPE_SHADER_FRAGMENT].dirty = false;
1054    }
1055 
1056    lima_update_textures(ctx);
1057 
1058    lima_pack_render_state(ctx, info);
1059    lima_pack_plbu_cmd(ctx, info, draw);
1060 
1061    if (ctx->gp_output) {
1062       lima_bo_unreference(ctx->gp_output); /* held by job */
1063       ctx->gp_output = NULL;
1064    }
1065 
1066    ctx->dirty = 0;
1067 }
1068 
1069 static void
lima_draw_vbo_indexed(struct pipe_context * pctx,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw)1070 lima_draw_vbo_indexed(struct pipe_context *pctx,
1071                       const struct pipe_draw_info *info,
1072                       const struct pipe_draw_start_count_bias *draw)
1073 {
1074    struct lima_context *ctx = lima_context(pctx);
1075    struct lima_job *job = lima_job_get(ctx);
1076    struct pipe_resource *indexbuf = NULL;
1077    bool needs_indices = true;
1078 
1079    /* Mali Utgard GPU always need min/max index info for index draw,
1080     * compute it if upper layer does not do for us */
1081    if (info->index_bounds_valid) {
1082       ctx->min_index = info->min_index;
1083       ctx->max_index = info->max_index;
1084       needs_indices = false;
1085    }
1086 
1087    if (info->has_user_indices) {
1088       util_upload_index_buffer(&ctx->base, info, draw, &indexbuf, &ctx->index_offset, 0x40);
1089       ctx->index_res = lima_resource(indexbuf);
1090    }
1091    else {
1092       ctx->index_res = lima_resource(info->index.resource);
1093       ctx->index_offset = 0;
1094       needs_indices = !panfrost_minmax_cache_get(ctx->index_res->index_cache, draw->start,
1095                                                  draw->count, &ctx->min_index, &ctx->max_index);
1096    }
1097 
1098    if (needs_indices) {
1099       u_vbuf_get_minmax_index(pctx, info, draw, &ctx->min_index, &ctx->max_index);
1100       if (!info->has_user_indices)
1101          panfrost_minmax_cache_add(ctx->index_res->index_cache, draw->start, draw->count,
1102                                    ctx->min_index, ctx->max_index);
1103    }
1104 
1105    lima_job_add_bo(job, LIMA_PIPE_GP, ctx->index_res->bo, LIMA_SUBMIT_BO_READ);
1106    lima_job_add_bo(job, LIMA_PIPE_PP, ctx->index_res->bo, LIMA_SUBMIT_BO_READ);
1107    lima_draw_vbo_update(pctx, info, draw);
1108 
1109    if (indexbuf)
1110       pipe_resource_reference(&indexbuf, NULL);
1111 }
1112 
1113 static void
lima_draw_vbo_count(struct pipe_context * pctx,const struct pipe_draw_info * info,const struct pipe_draw_start_count_bias * draw)1114 lima_draw_vbo_count(struct pipe_context *pctx,
1115                     const struct pipe_draw_info *info,
1116                     const struct pipe_draw_start_count_bias *draw)
1117 {
1118    static const uint32_t max_verts = 65535;
1119 
1120    struct pipe_draw_start_count_bias local_draw = *draw;
1121    unsigned start = draw->start;
1122    unsigned count = draw->count;
1123 
1124    while (count) {
1125       unsigned this_count = count;
1126       unsigned step;
1127 
1128       u_split_draw(info, max_verts, &this_count, &step);
1129 
1130       local_draw.start = start;
1131       local_draw.count = this_count;
1132 
1133       lima_draw_vbo_update(pctx, info, &local_draw);
1134 
1135       count -= step;
1136       start += step;
1137    }
1138 }
1139 
1140 static void
lima_draw_vbo(struct pipe_context * pctx,const struct pipe_draw_info * info,unsigned drawid_offset,const struct pipe_draw_indirect_info * indirect,const struct pipe_draw_start_count_bias * draws,unsigned num_draws)1141 lima_draw_vbo(struct pipe_context *pctx,
1142               const struct pipe_draw_info *info,
1143               unsigned drawid_offset,
1144               const struct pipe_draw_indirect_info *indirect,
1145               const struct pipe_draw_start_count_bias *draws,
1146               unsigned num_draws)
1147 {
1148    if (num_draws > 1) {
1149       util_draw_multi(pctx, info, drawid_offset, indirect, draws, num_draws);
1150       return;
1151    }
1152 
1153    /* check if draw mode and vertex/index count match,
1154     * otherwise gp will hang */
1155    if (!u_trim_pipe_prim(info->mode, (unsigned*)&draws[0].count)) {
1156       debug_printf("draw mode and vertex/index count mismatch\n");
1157       return;
1158    }
1159 
1160    struct lima_context *ctx = lima_context(pctx);
1161 
1162    if (!ctx->uncomp_fs || !ctx->uncomp_vs) {
1163       debug_warn_once("no shader, skip draw\n");
1164       return;
1165    }
1166 
1167    lima_clip_scissor_to_viewport(ctx);
1168    if (lima_is_scissor_zero(ctx))
1169       return;
1170 
1171    /* extend the viewport in case of line draws with a line_width > 1.0f,
1172     * otherwise use the original values */
1173    lima_extend_viewport(ctx, info);
1174 
1175    if (!lima_update_fs_state(ctx) || !lima_update_vs_state(ctx))
1176       return;
1177 
1178    struct lima_job *job = lima_job_get(ctx);
1179    job->pp_max_stack_size = MAX2(job->pp_max_stack_size, ctx->fs->state.stack_size);
1180 
1181    lima_dump_command_stream_print(
1182       job->dump, ctx->vs->bo->map, ctx->vs->state.shader_size, false,
1183       "add vs at va %x\n", ctx->vs->bo->va);
1184    lima_dump_shader(job->dump, ctx->vs->bo->map, ctx->vs->state.shader_size, false);
1185 
1186    lima_dump_command_stream_print(
1187       job->dump, ctx->fs->bo->map, ctx->fs->state.shader_size, false,
1188       "add fs at va %x\n", ctx->fs->bo->va);
1189    lima_dump_shader(job->dump, ctx->fs->bo->map, ctx->fs->state.shader_size, true);
1190 
1191    lima_job_add_bo(job, LIMA_PIPE_GP, ctx->vs->bo, LIMA_SUBMIT_BO_READ);
1192    lima_job_add_bo(job, LIMA_PIPE_PP, ctx->fs->bo, LIMA_SUBMIT_BO_READ);
1193 
1194    if (info->index_size)
1195       lima_draw_vbo_indexed(pctx, info, &draws[0]);
1196    else
1197       lima_draw_vbo_count(pctx, info, &draws[0]);
1198 
1199    job->draws++;
1200    /* Flush job if we hit the limit of draws per job otherwise we may
1201     * hit tile heap size limit */
1202    if (job->draws > MAX_DRAWS_PER_JOB) {
1203       unsigned resolve = job->resolve;
1204       lima_do_job(job);
1205       /* Subsequent job will need to resolve the same buffers */
1206       lima_update_job_wb(ctx, resolve);
1207    }
1208 }
1209 
1210 void
lima_draw_init(struct lima_context * ctx)1211 lima_draw_init(struct lima_context *ctx)
1212 {
1213    ctx->base.clear = lima_clear;
1214    ctx->base.draw_vbo = lima_draw_vbo;
1215 }
1216