1
/**************************************************************************
3
* Copyright 2007-2018 VMware, Inc.
6
* Permission is hereby granted, free of charge, to any person obtaining a
7
* copy of this software and associated documentation files (the
8
* "Software"), to deal in the Software without restriction, including
9
* without limitation the rights to use, copy, modify, merge, publish,
10
* distribute, sub license, and/or sell copies of the Software, and to
11
* permit persons to whom the Software is furnished to do so, subject to
12
* the following conditions:
14
* The above copyright notice and this permission notice (including the
15
* next paragraph) shall be included in all copies or substantial portions
18
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
19
* OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
20
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
21
* IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
22
* ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
23
* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
24
* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
26
**************************************************************************/
29
* AA line stage: AA lines are converted triangles (with extra generic)
35
#include "pipe/p_context.h"
36
#include "pipe/p_defines.h"
37
#include "pipe/p_shader_tokens.h"
38
#include "util/u_inlines.h"
40
#include "util/format/u_format.h"
41
#include "util/u_math.h"
42
#include "util/u_memory.h"
44
#include "tgsi/tgsi_transform.h"
45
#include "tgsi/tgsi_dump.h"
47
#include "draw_context.h"
48
#include "draw_private.h"
49
#include "draw_pipe.h"
52
#include "nir/nir_draw_helpers.h"
54
/** Approx number of new tokens for instructions in aa_transform_inst() */
55
#define NUM_NEW_TOKENS 53
59
* Subclass of pipe_shader_state to carry extra fragment shader info.
61
struct aaline_fragment_shader
63
struct pipe_shader_state state;
66
int generic_attrib; /**< generic used for distance */
71
* Subclass of draw_stage
75
struct draw_stage stage;
77
float half_line_width;
79
/** For AA lines, this is the vertex attrib slot for new generic */
81
/** position, not necessarily output zero */
86
* Currently bound state
88
struct aaline_fragment_shader *fs;
91
* Driver interface/override functions
93
void * (*driver_create_fs_state)(struct pipe_context *,
94
const struct pipe_shader_state *);
95
void (*driver_bind_fs_state)(struct pipe_context *, void *);
96
void (*driver_delete_fs_state)(struct pipe_context *, void *);
102
* Subclass of tgsi_transform_context, used for transforming the
103
* user's fragment shader to add the special AA instructions.
105
struct aa_transform_context {
106
struct tgsi_transform_context base;
107
uint64_t tempsUsed; /**< bitmask */
108
int colorOutput; /**< which output is the primary color */
109
int maxInput, maxGeneric; /**< max input index found */
110
int colorTemp, aaTemp; /**< temp registers */
114
* TGSI declaration transform callback.
115
* Look for a free input attrib, and two free temp regs.
118
aa_transform_decl(struct tgsi_transform_context *ctx,
119
struct tgsi_full_declaration *decl)
121
struct aa_transform_context *aactx = (struct aa_transform_context *)ctx;
123
if (decl->Declaration.File == TGSI_FILE_OUTPUT &&
124
decl->Semantic.Name == TGSI_SEMANTIC_COLOR &&
125
decl->Semantic.Index == 0) {
126
aactx->colorOutput = decl->Range.First;
128
else if (decl->Declaration.File == TGSI_FILE_INPUT) {
129
if ((int) decl->Range.Last > aactx->maxInput)
130
aactx->maxInput = decl->Range.Last;
131
if (decl->Semantic.Name == TGSI_SEMANTIC_GENERIC &&
132
(int) decl->Semantic.Index > aactx->maxGeneric) {
133
aactx->maxGeneric = decl->Semantic.Index;
136
else if (decl->Declaration.File == TGSI_FILE_TEMPORARY) {
138
for (i = decl->Range.First;
139
i <= decl->Range.Last; i++) {
141
* XXX this bitfield doesn't really cut it...
143
aactx->tempsUsed |= UINT64_C(1) << i;
147
ctx->emit_declaration(ctx, decl);
152
* Find the lowest zero bit, or -1 if bitfield is all ones.
155
free_bit(uint64_t bitfield)
157
return ffsll(~bitfield) - 1;
162
* TGSI transform prolog callback.
165
aa_transform_prolog(struct tgsi_transform_context *ctx)
167
struct aa_transform_context *aactx = (struct aa_transform_context *) ctx;
168
uint64_t usedTemps = aactx->tempsUsed;
170
/* find two free temp regs */
171
aactx->colorTemp = free_bit(usedTemps);
172
usedTemps |= UINT64_C(1) << aactx->colorTemp;
173
aactx->aaTemp = free_bit(usedTemps);
174
assert(aactx->colorTemp >= 0);
175
assert(aactx->aaTemp >= 0);
177
/* declare new generic input/texcoord */
178
tgsi_transform_input_decl(ctx, aactx->maxInput + 1,
179
TGSI_SEMANTIC_GENERIC, aactx->maxGeneric + 1,
180
TGSI_INTERPOLATE_LINEAR);
182
/* declare new temp regs */
183
tgsi_transform_temp_decl(ctx, aactx->aaTemp);
184
tgsi_transform_temp_decl(ctx, aactx->colorTemp);
189
* TGSI transform epilog callback.
192
aa_transform_epilog(struct tgsi_transform_context *ctx)
194
struct aa_transform_context *aactx = (struct aa_transform_context *) ctx;
196
if (aactx->colorOutput != -1) {
197
struct tgsi_full_instruction inst;
198
/* insert distance-based coverage code for antialiasing. */
200
/* saturate(linewidth - fabs(interpx), linelength - fabs(interpz) */
201
inst = tgsi_default_full_instruction();
202
inst.Instruction.Saturate = true;
203
inst.Instruction.Opcode = TGSI_OPCODE_ADD;
204
inst.Instruction.NumDstRegs = 1;
205
tgsi_transform_dst_reg(&inst.Dst[0], TGSI_FILE_TEMPORARY,
206
aactx->aaTemp, TGSI_WRITEMASK_XZ);
207
inst.Instruction.NumSrcRegs = 2;
208
tgsi_transform_src_reg(&inst.Src[1], TGSI_FILE_INPUT, aactx->maxInput + 1,
209
TGSI_SWIZZLE_X, TGSI_SWIZZLE_X,
210
TGSI_SWIZZLE_Z, TGSI_SWIZZLE_Z);
211
tgsi_transform_src_reg(&inst.Src[0], TGSI_FILE_INPUT, aactx->maxInput + 1,
212
TGSI_SWIZZLE_Y, TGSI_SWIZZLE_Y,
213
TGSI_SWIZZLE_W, TGSI_SWIZZLE_W);
214
inst.Src[1].Register.Absolute = true;
215
inst.Src[1].Register.Negate = true;
216
ctx->emit_instruction(ctx, &inst);
218
/* MUL width / height alpha */
219
tgsi_transform_op2_swz_inst(ctx, TGSI_OPCODE_MUL,
220
TGSI_FILE_TEMPORARY, aactx->aaTemp,
222
TGSI_FILE_TEMPORARY, aactx->aaTemp,
224
TGSI_FILE_TEMPORARY, aactx->aaTemp,
225
TGSI_SWIZZLE_Z, false);
228
tgsi_transform_op1_inst(ctx, TGSI_OPCODE_MOV,
229
TGSI_FILE_OUTPUT, aactx->colorOutput,
231
TGSI_FILE_TEMPORARY, aactx->colorTemp);
234
tgsi_transform_op2_inst(ctx, TGSI_OPCODE_MUL,
235
TGSI_FILE_OUTPUT, aactx->colorOutput,
237
TGSI_FILE_TEMPORARY, aactx->colorTemp,
238
TGSI_FILE_TEMPORARY, aactx->aaTemp, false);
244
* TGSI instruction transform callback.
245
* Replace writes to result.color w/ a temp reg.
248
aa_transform_inst(struct tgsi_transform_context *ctx,
249
struct tgsi_full_instruction *inst)
251
struct aa_transform_context *aactx = (struct aa_transform_context *) ctx;
255
* Look for writes to result.color and replace with colorTemp reg.
257
for (i = 0; i < inst->Instruction.NumDstRegs; i++) {
258
struct tgsi_full_dst_register *dst = &inst->Dst[i];
259
if (dst->Register.File == TGSI_FILE_OUTPUT &&
260
dst->Register.Index == aactx->colorOutput) {
261
dst->Register.File = TGSI_FILE_TEMPORARY;
262
dst->Register.Index = aactx->colorTemp;
266
ctx->emit_instruction(ctx, inst);
271
* Generate the frag shader we'll use for drawing AA lines.
272
* This will be the user's shader plus some arithmetic instructions.
275
generate_aaline_fs(struct aaline_stage *aaline)
277
struct pipe_context *pipe = aaline->stage.draw->pipe;
278
const struct pipe_shader_state *orig_fs = &aaline->fs->state;
279
struct pipe_shader_state aaline_fs;
280
struct aa_transform_context transform;
281
const uint newLen = tgsi_num_tokens(orig_fs->tokens) + NUM_NEW_TOKENS;
283
aaline_fs = *orig_fs; /* copy to init */
285
memset(&transform, 0, sizeof(transform));
286
transform.colorOutput = -1;
287
transform.maxInput = -1;
288
transform.maxGeneric = -1;
289
transform.colorTemp = -1;
290
transform.aaTemp = -1;
291
transform.base.prolog = aa_transform_prolog;
292
transform.base.epilog = aa_transform_epilog;
293
transform.base.transform_instruction = aa_transform_inst;
294
transform.base.transform_declaration = aa_transform_decl;
296
aaline_fs.tokens = tgsi_transform_shader(orig_fs->tokens, newLen, &transform.base);
297
if (!aaline_fs.tokens)
301
debug_printf("draw_aaline, orig shader:\n");
302
tgsi_dump(orig_fs->tokens, 0);
303
debug_printf("draw_aaline, new shader:\n");
304
tgsi_dump(aaline_fs.tokens, 0);
307
aaline->fs->aaline_fs = aaline->driver_create_fs_state(pipe, &aaline_fs);
308
if (aaline->fs->aaline_fs != NULL)
309
aaline->fs->generic_attrib = transform.maxGeneric + 1;
311
FREE((void *)aaline_fs.tokens);
312
return aaline->fs->aaline_fs != NULL;
316
generate_aaline_fs_nir(struct aaline_stage *aaline)
318
struct pipe_context *pipe = aaline->stage.draw->pipe;
319
const struct pipe_shader_state *orig_fs = &aaline->fs->state;
320
struct pipe_shader_state aaline_fs;
322
aaline_fs = *orig_fs; /* copy to init */
323
aaline_fs.ir.nir = nir_shader_clone(NULL, orig_fs->ir.nir);
324
if (!aaline_fs.ir.nir)
327
nir_lower_aaline_fs(aaline_fs.ir.nir, &aaline->fs->generic_attrib);
328
aaline->fs->aaline_fs = aaline->driver_create_fs_state(pipe, &aaline_fs);
329
if (aaline->fs->aaline_fs == NULL)
336
* When we're about to draw our first AA line in a batch, this function is
337
* called to tell the driver to bind our modified fragment shader.
340
bind_aaline_fragment_shader(struct aaline_stage *aaline)
342
struct draw_context *draw = aaline->stage.draw;
343
struct pipe_context *pipe = draw->pipe;
345
if (!aaline->fs->aaline_fs) {
346
if (aaline->fs->state.type == PIPE_SHADER_IR_NIR) {
347
if (!generate_aaline_fs_nir(aaline))
350
if (!generate_aaline_fs(aaline))
354
draw->suspend_flushing = TRUE;
355
aaline->driver_bind_fs_state(pipe, aaline->fs->aaline_fs);
356
draw->suspend_flushing = FALSE;
363
static inline struct aaline_stage *
364
aaline_stage(struct draw_stage *stage)
366
return (struct aaline_stage *) stage;
371
* Draw a wide line by drawing a quad, using geometry which will
372
* fullfill GL's antialiased line requirements.
375
aaline_line(struct draw_stage *stage, struct prim_header *header)
377
const struct aaline_stage *aaline = aaline_stage(stage);
378
const float half_width = aaline->half_line_width;
379
struct prim_header tri;
380
struct vertex_header *v[8];
381
uint coordPos = aaline->coord_slot;
382
uint posPos = aaline->pos_slot;
384
float dx = header->v[1]->data[posPos][0] - header->v[0]->data[posPos][0];
385
float dy = header->v[1]->data[posPos][1] - header->v[0]->data[posPos][1];
386
float a = atan2f(dy, dx);
387
float c_a = cosf(a), s_a = sinf(a);
392
half_length = 0.5f * sqrtf(dx * dx + dy * dy);
394
if (half_length < 0.5f) {
396
* The logic we use for "normal" sized segments is incorrect
397
* for very short segments (basically because we only have
398
* one value to interpolate, not a distance to each endpoint).
399
* Therefore, we calculate half_length differently, so that for
400
* original line length (near) 0, we get alpha 0 - otherwise
401
* max alpha would still be 0.5. This also prevents us from
402
* artifacts due to degenerated lines (the endpoints being
403
* identical, which would still receive anywhere from alpha
404
* 0-0.5 otherwise) (at least the pstipple stage may generate
405
* such lines due to float inaccuracies if line length is very
406
* close to a integer).
407
* Might not be fully accurate neither (because the "strength" of
408
* the line is going to be determined by how close to the pixel
409
* center those 1 or 2 fragments are) but it's probably the best
412
half_length = 2.0f * half_length;
414
half_length = half_length + 0.5f;
420
/* allocate/dup new verts */
421
for (i = 0; i < 4; i++) {
422
v[i] = dup_vert(stage, header->v[i/2], i);
426
* Quad strip for line from v0 to v1 (*=endpoints):
429
* +-----------------------------+
433
* +-----------------------------+
438
* We increase line length by 0.5 pixels (at each endpoint),
439
* and calculate the tri endpoints by moving them half-width
440
* distance away perpendicular to the line.
441
* XXX: since we change line endpoints (by 0.5 pixel), should
442
* actually re-interpolate all other values?
446
pos = v[0]->data[posPos];
447
pos[0] += (-t_l * c_a - t_w * s_a);
448
pos[1] += (-t_l * s_a + t_w * c_a);
450
pos = v[1]->data[posPos];
451
pos[0] += (-t_l * c_a - -t_w * s_a);
452
pos[1] += (-t_l * s_a + -t_w * c_a);
454
pos = v[2]->data[posPos];
455
pos[0] += (t_l * c_a - t_w * s_a);
456
pos[1] += (t_l * s_a + t_w * c_a);
458
pos = v[3]->data[posPos];
459
pos[0] += (t_l * c_a - -t_w * s_a);
460
pos[1] += (t_l * s_a + -t_w * c_a);
463
tex = v[0]->data[coordPos];
464
ASSIGN_4V(tex, -half_width, half_width, -half_length, half_length);
466
tex = v[1]->data[coordPos];
467
ASSIGN_4V(tex, half_width, half_width, -half_length, half_length);
469
tex = v[2]->data[coordPos];
470
ASSIGN_4V(tex, -half_width, half_width, half_length, half_length);
472
tex = v[3]->data[coordPos];
473
ASSIGN_4V(tex, half_width, half_width, half_length, half_length);
475
tri.v[0] = v[2]; tri.v[1] = v[1]; tri.v[2] = v[0];
476
stage->next->tri(stage->next, &tri);
478
tri.v[0] = v[3]; tri.v[1] = v[1]; tri.v[2] = v[2];
479
stage->next->tri(stage->next, &tri);
484
aaline_first_line(struct draw_stage *stage, struct prim_header *header)
486
auto struct aaline_stage *aaline = aaline_stage(stage);
487
struct draw_context *draw = stage->draw;
488
struct pipe_context *pipe = draw->pipe;
489
const struct pipe_rasterizer_state *rast = draw->rasterizer;
492
assert(draw->rasterizer->line_smooth && !draw->rasterizer->multisample);
494
if (draw->rasterizer->line_width <= 1.0)
495
aaline->half_line_width = 1.0;
497
aaline->half_line_width = 0.5f * draw->rasterizer->line_width + 0.5f;
499
if (!draw->rasterizer->half_pixel_center)
501
* The tex coords probably would need adjustments?
503
debug_printf("aa lines without half pixel center may be wrong\n");
506
* Bind (generate) our fragprog
508
if (!bind_aaline_fragment_shader(aaline)) {
509
stage->line = draw_pipe_passthrough_line;
510
stage->line(stage, header);
514
draw_aaline_prepare_outputs(draw, draw->pipeline.aaline);
516
draw->suspend_flushing = TRUE;
518
/* Disable triangle culling, stippling, unfilled mode etc. */
519
r = draw_get_rasterizer_no_cull(draw, rast);
520
pipe->bind_rasterizer_state(pipe, r);
522
draw->suspend_flushing = FALSE;
524
/* now really draw first line */
525
stage->line = aaline_line;
526
stage->line(stage, header);
531
aaline_flush(struct draw_stage *stage, unsigned flags)
533
struct draw_context *draw = stage->draw;
534
struct aaline_stage *aaline = aaline_stage(stage);
535
struct pipe_context *pipe = draw->pipe;
537
stage->line = aaline_first_line;
538
stage->next->flush(stage->next, flags);
540
/* restore original frag shader */
541
draw->suspend_flushing = TRUE;
542
aaline->driver_bind_fs_state(pipe, aaline->fs ? aaline->fs->driver_fs : NULL);
544
/* restore original rasterizer state */
545
if (draw->rast_handle) {
546
pipe->bind_rasterizer_state(pipe, draw->rast_handle);
549
draw->suspend_flushing = FALSE;
551
draw_remove_extra_vertex_attribs(draw);
556
aaline_reset_stipple_counter(struct draw_stage *stage)
558
stage->next->reset_stipple_counter(stage->next);
563
aaline_destroy(struct draw_stage *stage)
565
struct aaline_stage *aaline = aaline_stage(stage);
566
struct pipe_context *pipe = stage->draw->pipe;
568
draw_free_temp_verts(stage);
570
/* restore the old entry points */
571
pipe->create_fs_state = aaline->driver_create_fs_state;
572
pipe->bind_fs_state = aaline->driver_bind_fs_state;
573
pipe->delete_fs_state = aaline->driver_delete_fs_state;
579
static struct aaline_stage *
580
draw_aaline_stage(struct draw_context *draw)
582
struct aaline_stage *aaline = CALLOC_STRUCT(aaline_stage);
586
aaline->stage.draw = draw;
587
aaline->stage.name = "aaline";
588
aaline->stage.next = NULL;
589
aaline->stage.point = draw_pipe_passthrough_point;
590
aaline->stage.line = aaline_first_line;
591
aaline->stage.tri = draw_pipe_passthrough_tri;
592
aaline->stage.flush = aaline_flush;
593
aaline->stage.reset_stipple_counter = aaline_reset_stipple_counter;
594
aaline->stage.destroy = aaline_destroy;
596
if (!draw_alloc_temp_verts(&aaline->stage, 8)) {
597
aaline->stage.destroy(&aaline->stage);
605
static struct aaline_stage *
606
aaline_stage_from_pipe(struct pipe_context *pipe)
608
struct draw_context *draw = (struct draw_context *) pipe->draw;
611
return aaline_stage(draw->pipeline.aaline);
619
* This function overrides the driver's create_fs_state() function and
620
* will typically be called by the gallium frontend.
623
aaline_create_fs_state(struct pipe_context *pipe,
624
const struct pipe_shader_state *fs)
626
struct aaline_stage *aaline = aaline_stage_from_pipe(pipe);
627
struct aaline_fragment_shader *aafs = NULL;
632
aafs = CALLOC_STRUCT(aaline_fragment_shader);
637
aafs->state.type = fs->type;
638
if (fs->type == PIPE_SHADER_IR_TGSI)
639
aafs->state.tokens = tgsi_dup_tokens(fs->tokens);
641
aafs->state.ir.nir = nir_shader_clone(NULL, fs->ir.nir);
644
aafs->driver_fs = aaline->driver_create_fs_state(pipe, fs);
651
aaline_bind_fs_state(struct pipe_context *pipe, void *fs)
653
struct aaline_stage *aaline = aaline_stage_from_pipe(pipe);
654
struct aaline_fragment_shader *aafs = (struct aaline_fragment_shader *) fs;
663
aaline->driver_bind_fs_state(pipe, (aafs ? aafs->driver_fs : NULL));
668
aaline_delete_fs_state(struct pipe_context *pipe, void *fs)
670
struct aaline_stage *aaline = aaline_stage_from_pipe(pipe);
671
struct aaline_fragment_shader *aafs = (struct aaline_fragment_shader *) fs;
679
aaline->driver_delete_fs_state(pipe, aafs->driver_fs);
682
aaline->driver_delete_fs_state(pipe, aafs->aaline_fs);
685
if (aafs->state.type == PIPE_SHADER_IR_TGSI)
686
FREE((void*)aafs->state.tokens);
688
ralloc_free(aafs->state.ir.nir);
694
draw_aaline_prepare_outputs(struct draw_context *draw,
695
struct draw_stage *stage)
697
struct aaline_stage *aaline = aaline_stage(stage);
698
const struct pipe_rasterizer_state *rast = draw->rasterizer;
700
/* update vertex attrib info */
701
aaline->pos_slot = draw_current_shader_position_output(draw);
703
if (!rast->line_smooth || rast->multisample)
706
/* allocate the extra post-transformed vertex attribute */
707
if (aaline->fs && aaline->fs->aaline_fs)
708
aaline->coord_slot = draw_alloc_extra_vertex_attrib(draw,
709
TGSI_SEMANTIC_GENERIC,
710
aaline->fs->generic_attrib);
712
aaline->coord_slot = -1;
716
* Called by drivers that want to install this AA line prim stage
717
* into the draw module's pipeline. This will not be used if the
718
* hardware has native support for AA lines.
721
draw_install_aaline_stage(struct draw_context *draw, struct pipe_context *pipe)
723
struct aaline_stage *aaline;
725
pipe->draw = (void *) draw;
728
* Create / install AA line drawing / prim stage
730
aaline = draw_aaline_stage(draw);
734
/* save original driver functions */
735
aaline->driver_create_fs_state = pipe->create_fs_state;
736
aaline->driver_bind_fs_state = pipe->bind_fs_state;
737
aaline->driver_delete_fs_state = pipe->delete_fs_state;
739
/* override the driver's functions */
740
pipe->create_fs_state = aaline_create_fs_state;
741
pipe->bind_fs_state = aaline_bind_fs_state;
742
pipe->delete_fs_state = aaline_delete_fs_state;
744
/* Install once everything is known to be OK:
746
draw->pipeline.aaline = &aaline->stage;