2 * Copyright (C) 2012 Rob Clark <robclark@freedesktop.org>
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:
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
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 FROM,
20 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
24 * Rob Clark <robclark@freedesktop.org>
27 #include "pipe/p_state.h"
28 #include "util/u_dual_blend.h"
29 #include "util/u_string.h"
30 #include "util/u_memory.h"
31 #include "util/u_helpers.h"
33 #include "freedreno_state.h"
34 #include "freedreno_context.h"
35 #include "freedreno_resource.h"
36 #include "freedreno_texture.h"
37 #include "freedreno_gmem.h"
38 #include "freedreno_query_hw.h"
39 #include "freedreno_util.h"
41 /* All the generic state handling.. In case of CSO's that are specific
42 * to the GPU version, when the bind and the delete are common they can
47 fd_set_blend_color(struct pipe_context
*pctx
,
48 const struct pipe_blend_color
*blend_color
)
50 struct fd_context
*ctx
= fd_context(pctx
);
51 ctx
->blend_color
= *blend_color
;
52 ctx
->dirty
|= FD_DIRTY_BLEND_COLOR
;
56 fd_set_stencil_ref(struct pipe_context
*pctx
,
57 const struct pipe_stencil_ref
*stencil_ref
)
59 struct fd_context
*ctx
= fd_context(pctx
);
60 ctx
->stencil_ref
=* stencil_ref
;
61 ctx
->dirty
|= FD_DIRTY_STENCIL_REF
;
65 fd_set_clip_state(struct pipe_context
*pctx
,
66 const struct pipe_clip_state
*clip
)
68 struct fd_context
*ctx
= fd_context(pctx
);
70 ctx
->dirty
|= FD_DIRTY_UCP
;
74 fd_set_sample_mask(struct pipe_context
*pctx
, unsigned sample_mask
)
76 struct fd_context
*ctx
= fd_context(pctx
);
77 ctx
->sample_mask
= (uint16_t)sample_mask
;
78 ctx
->dirty
|= FD_DIRTY_SAMPLE_MASK
;
82 fd_set_min_samples(struct pipe_context
*pctx
, unsigned min_samples
)
84 struct fd_context
*ctx
= fd_context(pctx
);
85 ctx
->min_samples
= min_samples
;
86 ctx
->dirty
|= FD_DIRTY_MIN_SAMPLES
;
89 /* notes from calim on #dri-devel:
90 * index==0 will be non-UBO (ie. glUniformXYZ()) all packed together padded
92 * I should be able to consider that I own the user_ptr until the next
93 * set_constant_buffer() call, at which point I don't really care about the
95 * index>0 will be UBO's.. well, I'll worry about that later
98 fd_set_constant_buffer(struct pipe_context
*pctx
,
99 enum pipe_shader_type shader
, uint index
,
100 const struct pipe_constant_buffer
*cb
)
102 struct fd_context
*ctx
= fd_context(pctx
);
103 struct fd_constbuf_stateobj
*so
= &ctx
->constbuf
[shader
];
105 util_copy_constant_buffer(&so
->cb
[index
], cb
);
107 /* Note that the state tracker can unbind constant buffers by
111 so
->enabled_mask
&= ~(1 << index
);
115 so
->enabled_mask
|= 1 << index
;
116 ctx
->dirty_shader
[shader
] |= FD_DIRTY_SHADER_CONST
;
117 ctx
->dirty
|= FD_DIRTY_CONST
;
121 fd_set_shader_buffers(struct pipe_context
*pctx
,
122 enum pipe_shader_type shader
,
123 unsigned start
, unsigned count
,
124 const struct pipe_shader_buffer
*buffers
,
125 unsigned writable_bitmask
)
127 struct fd_context
*ctx
= fd_context(pctx
);
128 struct fd_shaderbuf_stateobj
*so
= &ctx
->shaderbuf
[shader
];
132 for (unsigned i
= 0; i
< count
; i
++) {
133 unsigned n
= i
+ start
;
134 struct pipe_shader_buffer
*buf
= &so
->sb
[n
];
136 if ((buf
->buffer
== buffers
[i
].buffer
) &&
137 (buf
->buffer_offset
== buffers
[i
].buffer_offset
) &&
138 (buf
->buffer_size
== buffers
[i
].buffer_size
))
143 buf
->buffer_offset
= buffers
[i
].buffer_offset
;
144 buf
->buffer_size
= buffers
[i
].buffer_size
;
145 pipe_resource_reference(&buf
->buffer
, buffers
[i
].buffer
);
148 so
->enabled_mask
|= BIT(n
);
150 so
->enabled_mask
&= ~BIT(n
);
153 mask
= (BIT(count
) - 1) << start
;
155 for (unsigned i
= 0; i
< count
; i
++) {
156 unsigned n
= i
+ start
;
157 struct pipe_shader_buffer
*buf
= &so
->sb
[n
];
159 pipe_resource_reference(&buf
->buffer
, NULL
);
162 so
->enabled_mask
&= ~mask
;
165 ctx
->dirty_shader
[shader
] |= FD_DIRTY_SHADER_SSBO
;
169 fd_set_shader_images(struct pipe_context
*pctx
,
170 enum pipe_shader_type shader
,
171 unsigned start
, unsigned count
,
172 const struct pipe_image_view
*images
)
174 struct fd_context
*ctx
= fd_context(pctx
);
175 struct fd_shaderimg_stateobj
*so
= &ctx
->shaderimg
[shader
];
180 for (unsigned i
= 0; i
< count
; i
++) {
181 unsigned n
= i
+ start
;
182 struct pipe_image_view
*buf
= &so
->si
[n
];
184 if ((buf
->resource
== images
[i
].resource
) &&
185 (buf
->format
== images
[i
].format
) &&
186 (buf
->access
== images
[i
].access
) &&
187 !memcmp(&buf
->u
, &images
[i
].u
, sizeof(buf
->u
)))
191 util_copy_image_view(buf
, &images
[i
]);
194 so
->enabled_mask
|= BIT(n
);
196 so
->enabled_mask
&= ~BIT(n
);
199 mask
= (BIT(count
) - 1) << start
;
201 for (unsigned i
= 0; i
< count
; i
++) {
202 unsigned n
= i
+ start
;
203 struct pipe_image_view
*img
= &so
->si
[n
];
205 pipe_resource_reference(&img
->resource
, NULL
);
208 so
->enabled_mask
&= ~mask
;
211 ctx
->dirty_shader
[shader
] |= FD_DIRTY_SHADER_IMAGE
;
215 fd_set_framebuffer_state(struct pipe_context
*pctx
,
216 const struct pipe_framebuffer_state
*framebuffer
)
218 struct fd_context
*ctx
= fd_context(pctx
);
219 struct pipe_framebuffer_state
*cso
;
221 DBG("%ux%u, %u layers, %u samples",
222 framebuffer
->width
, framebuffer
->height
,
223 framebuffer
->layers
, framebuffer
->samples
);
225 cso
= &ctx
->framebuffer
;
227 if (util_framebuffer_state_equal(cso
, framebuffer
))
230 util_copy_framebuffer_state(cso
, framebuffer
);
232 cso
->samples
= util_framebuffer_get_num_samples(cso
);
234 if (ctx
->screen
->reorder
) {
235 struct fd_batch
*old_batch
= NULL
;
237 fd_batch_reference(&old_batch
, ctx
->batch
);
239 if (likely(old_batch
))
240 fd_batch_set_stage(old_batch
, FD_STAGE_NULL
);
242 fd_batch_reference(&ctx
->batch
, NULL
);
243 fd_context_all_dirty(ctx
);
245 if (old_batch
&& old_batch
->blit
&& !old_batch
->back_blit
) {
246 /* for blits, there is not really much point in hanging on
247 * to the uncommitted batch (ie. you probably don't blit
248 * multiple times to the same surface), so we might as
249 * well go ahead and flush this one:
251 fd_batch_flush(old_batch
, false);
254 fd_batch_reference(&old_batch
, NULL
);
256 DBG("%d: cbufs[0]=%p, zsbuf=%p", ctx
->batch
->needs_flush
,
257 framebuffer
->cbufs
[0], framebuffer
->zsbuf
);
258 fd_batch_flush(ctx
->batch
, false);
259 util_copy_framebuffer_state(&ctx
->batch
->framebuffer
, cso
);
262 ctx
->dirty
|= FD_DIRTY_FRAMEBUFFER
;
264 ctx
->disabled_scissor
.minx
= 0;
265 ctx
->disabled_scissor
.miny
= 0;
266 ctx
->disabled_scissor
.maxx
= cso
->width
;
267 ctx
->disabled_scissor
.maxy
= cso
->height
;
269 ctx
->dirty
|= FD_DIRTY_SCISSOR
;
273 fd_set_polygon_stipple(struct pipe_context
*pctx
,
274 const struct pipe_poly_stipple
*stipple
)
276 struct fd_context
*ctx
= fd_context(pctx
);
277 ctx
->stipple
= *stipple
;
278 ctx
->dirty
|= FD_DIRTY_STIPPLE
;
282 fd_set_scissor_states(struct pipe_context
*pctx
,
284 unsigned num_scissors
,
285 const struct pipe_scissor_state
*scissor
)
287 struct fd_context
*ctx
= fd_context(pctx
);
289 ctx
->scissor
= *scissor
;
290 ctx
->dirty
|= FD_DIRTY_SCISSOR
;
294 fd_set_viewport_states(struct pipe_context
*pctx
,
296 unsigned num_viewports
,
297 const struct pipe_viewport_state
*viewport
)
299 struct fd_context
*ctx
= fd_context(pctx
);
300 struct pipe_scissor_state
*scissor
= &ctx
->viewport_scissor
;
301 float minx
, miny
, maxx
, maxy
;
303 ctx
->viewport
= *viewport
;
305 /* see si_get_scissor_from_viewport(): */
307 /* Convert (-1, -1) and (1, 1) from clip space into window space. */
308 minx
= -viewport
->scale
[0] + viewport
->translate
[0];
309 miny
= -viewport
->scale
[1] + viewport
->translate
[1];
310 maxx
= viewport
->scale
[0] + viewport
->translate
[0];
311 maxy
= viewport
->scale
[1] + viewport
->translate
[1];
313 /* Handle inverted viewports. */
321 debug_assert(miny
>= 0);
322 debug_assert(maxy
>= 0);
324 /* Convert to integer and round up the max bounds. */
325 scissor
->minx
= minx
;
326 scissor
->miny
= miny
;
327 scissor
->maxx
= ceilf(maxx
);
328 scissor
->maxy
= ceilf(maxy
);
330 ctx
->dirty
|= FD_DIRTY_VIEWPORT
;
334 fd_set_vertex_buffers(struct pipe_context
*pctx
,
335 unsigned start_slot
, unsigned count
,
336 const struct pipe_vertex_buffer
*vb
)
338 struct fd_context
*ctx
= fd_context(pctx
);
339 struct fd_vertexbuf_stateobj
*so
= &ctx
->vtx
.vertexbuf
;
342 /* on a2xx, pitch is encoded in the vtx fetch instruction, so
343 * we need to mark VTXSTATE as dirty as well to trigger patching
344 * and re-emitting the vtx shader:
346 if (ctx
->screen
->gpu_id
< 300) {
347 for (i
= 0; i
< count
; i
++) {
348 bool new_enabled
= vb
&& vb
[i
].buffer
.resource
;
349 bool old_enabled
= so
->vb
[i
].buffer
.resource
!= NULL
;
350 uint32_t new_stride
= vb
? vb
[i
].stride
: 0;
351 uint32_t old_stride
= so
->vb
[i
].stride
;
352 if ((new_enabled
!= old_enabled
) || (new_stride
!= old_stride
)) {
353 ctx
->dirty
|= FD_DIRTY_VTXSTATE
;
359 util_set_vertex_buffers_mask(so
->vb
, &so
->enabled_mask
, vb
, start_slot
, count
);
360 so
->count
= util_last_bit(so
->enabled_mask
);
362 ctx
->dirty
|= FD_DIRTY_VTXBUF
;
366 fd_blend_state_bind(struct pipe_context
*pctx
, void *hwcso
)
368 struct fd_context
*ctx
= fd_context(pctx
);
369 struct pipe_blend_state
*cso
= hwcso
;
370 bool old_is_dual
= ctx
->blend
?
371 ctx
->blend
->rt
[0].blend_enable
&& util_blend_state_is_dual(ctx
->blend
, 0) :
373 bool new_is_dual
= cso
?
374 cso
->rt
[0].blend_enable
&& util_blend_state_is_dual(cso
, 0) :
377 ctx
->dirty
|= FD_DIRTY_BLEND
;
378 if (old_is_dual
!= new_is_dual
)
379 ctx
->dirty
|= FD_DIRTY_BLEND_DUAL
;
383 fd_blend_state_delete(struct pipe_context
*pctx
, void *hwcso
)
389 fd_rasterizer_state_bind(struct pipe_context
*pctx
, void *hwcso
)
391 struct fd_context
*ctx
= fd_context(pctx
);
392 struct pipe_scissor_state
*old_scissor
= fd_context_get_scissor(ctx
);
394 ctx
->rasterizer
= hwcso
;
395 ctx
->dirty
|= FD_DIRTY_RASTERIZER
;
397 /* if scissor enable bit changed we need to mark scissor
398 * state as dirty as well:
399 * NOTE: we can do a shallow compare, since we only care
400 * if it changed to/from &ctx->disable_scissor
402 if (old_scissor
!= fd_context_get_scissor(ctx
))
403 ctx
->dirty
|= FD_DIRTY_SCISSOR
;
407 fd_rasterizer_state_delete(struct pipe_context
*pctx
, void *hwcso
)
413 fd_zsa_state_bind(struct pipe_context
*pctx
, void *hwcso
)
415 struct fd_context
*ctx
= fd_context(pctx
);
417 ctx
->dirty
|= FD_DIRTY_ZSA
;
421 fd_zsa_state_delete(struct pipe_context
*pctx
, void *hwcso
)
427 fd_vertex_state_create(struct pipe_context
*pctx
, unsigned num_elements
,
428 const struct pipe_vertex_element
*elements
)
430 struct fd_vertex_stateobj
*so
= CALLOC_STRUCT(fd_vertex_stateobj
);
435 memcpy(so
->pipe
, elements
, sizeof(*elements
) * num_elements
);
436 so
->num_elements
= num_elements
;
442 fd_vertex_state_delete(struct pipe_context
*pctx
, void *hwcso
)
448 fd_vertex_state_bind(struct pipe_context
*pctx
, void *hwcso
)
450 struct fd_context
*ctx
= fd_context(pctx
);
451 ctx
->vtx
.vtx
= hwcso
;
452 ctx
->dirty
|= FD_DIRTY_VTXSTATE
;
455 static struct pipe_stream_output_target
*
456 fd_create_stream_output_target(struct pipe_context
*pctx
,
457 struct pipe_resource
*prsc
, unsigned buffer_offset
,
458 unsigned buffer_size
)
460 struct pipe_stream_output_target
*target
;
461 struct fd_resource
*rsc
= fd_resource(prsc
);
463 target
= CALLOC_STRUCT(pipe_stream_output_target
);
467 pipe_reference_init(&target
->reference
, 1);
468 pipe_resource_reference(&target
->buffer
, prsc
);
470 target
->context
= pctx
;
471 target
->buffer_offset
= buffer_offset
;
472 target
->buffer_size
= buffer_size
;
474 assert(rsc
->base
.target
== PIPE_BUFFER
);
475 util_range_add(&rsc
->valid_buffer_range
,
476 buffer_offset
, buffer_offset
+ buffer_size
);
482 fd_stream_output_target_destroy(struct pipe_context
*pctx
,
483 struct pipe_stream_output_target
*target
)
485 pipe_resource_reference(&target
->buffer
, NULL
);
490 fd_set_stream_output_targets(struct pipe_context
*pctx
,
491 unsigned num_targets
, struct pipe_stream_output_target
**targets
,
492 const unsigned *offsets
)
494 struct fd_context
*ctx
= fd_context(pctx
);
495 struct fd_streamout_stateobj
*so
= &ctx
->streamout
;
498 debug_assert(num_targets
<= ARRAY_SIZE(so
->targets
));
500 for (i
= 0; i
< num_targets
; i
++) {
501 boolean changed
= targets
[i
] != so
->targets
[i
];
502 boolean reset
= (offsets
[i
] != (unsigned)-1);
504 so
->reset
|= (reset
<< i
);
506 if (!changed
&& !reset
)
510 so
->offsets
[i
] = offsets
[i
];
512 pipe_so_target_reference(&so
->targets
[i
], targets
[i
]);
515 for (; i
< so
->num_targets
; i
++) {
516 pipe_so_target_reference(&so
->targets
[i
], NULL
);
519 so
->num_targets
= num_targets
;
521 ctx
->dirty
|= FD_DIRTY_STREAMOUT
;
525 fd_bind_compute_state(struct pipe_context
*pctx
, void *state
)
527 struct fd_context
*ctx
= fd_context(pctx
);
528 ctx
->compute
= state
;
529 ctx
->dirty_shader
[PIPE_SHADER_COMPUTE
] |= FD_DIRTY_SHADER_PROG
;
533 fd_set_compute_resources(struct pipe_context
*pctx
,
534 unsigned start
, unsigned count
, struct pipe_surface
**prscs
)
539 /* used by clover to bind global objects, returning the bo address
543 fd_set_global_binding(struct pipe_context
*pctx
,
544 unsigned first
, unsigned count
, struct pipe_resource
**prscs
,
547 struct fd_context
*ctx
= fd_context(pctx
);
548 struct fd_global_bindings_stateobj
*so
= &ctx
->global_bindings
;
552 for (unsigned i
= 0; i
< count
; i
++) {
553 unsigned n
= i
+ first
;
557 pipe_resource_reference(&so
->buf
[n
], prscs
[i
]);
560 struct fd_resource
*rsc
= fd_resource(so
->buf
[n
]);
561 uint64_t iova
= fd_bo_get_iova(rsc
->bo
);
562 // TODO need to scream if iova > 32b or fix gallium API..
567 so
->enabled_mask
|= BIT(n
);
569 so
->enabled_mask
&= ~BIT(n
);
572 mask
= (BIT(count
) - 1) << first
;
574 for (unsigned i
= 0; i
< count
; i
++) {
575 unsigned n
= i
+ first
;
577 struct fd_resource
*rsc
= fd_resource(so
->buf
[n
]);
578 fd_bo_put_iova(rsc
->bo
);
580 pipe_resource_reference(&so
->buf
[n
], NULL
);
583 so
->enabled_mask
&= ~mask
;
589 fd_state_init(struct pipe_context
*pctx
)
591 pctx
->set_blend_color
= fd_set_blend_color
;
592 pctx
->set_stencil_ref
= fd_set_stencil_ref
;
593 pctx
->set_clip_state
= fd_set_clip_state
;
594 pctx
->set_sample_mask
= fd_set_sample_mask
;
595 pctx
->set_min_samples
= fd_set_min_samples
;
596 pctx
->set_constant_buffer
= fd_set_constant_buffer
;
597 pctx
->set_shader_buffers
= fd_set_shader_buffers
;
598 pctx
->set_shader_images
= fd_set_shader_images
;
599 pctx
->set_framebuffer_state
= fd_set_framebuffer_state
;
600 pctx
->set_polygon_stipple
= fd_set_polygon_stipple
;
601 pctx
->set_scissor_states
= fd_set_scissor_states
;
602 pctx
->set_viewport_states
= fd_set_viewport_states
;
604 pctx
->set_vertex_buffers
= fd_set_vertex_buffers
;
606 pctx
->bind_blend_state
= fd_blend_state_bind
;
607 pctx
->delete_blend_state
= fd_blend_state_delete
;
609 pctx
->bind_rasterizer_state
= fd_rasterizer_state_bind
;
610 pctx
->delete_rasterizer_state
= fd_rasterizer_state_delete
;
612 pctx
->bind_depth_stencil_alpha_state
= fd_zsa_state_bind
;
613 pctx
->delete_depth_stencil_alpha_state
= fd_zsa_state_delete
;
615 pctx
->create_vertex_elements_state
= fd_vertex_state_create
;
616 pctx
->delete_vertex_elements_state
= fd_vertex_state_delete
;
617 pctx
->bind_vertex_elements_state
= fd_vertex_state_bind
;
619 pctx
->create_stream_output_target
= fd_create_stream_output_target
;
620 pctx
->stream_output_target_destroy
= fd_stream_output_target_destroy
;
621 pctx
->set_stream_output_targets
= fd_set_stream_output_targets
;
623 if (has_compute(fd_screen(pctx
->screen
))) {
624 pctx
->bind_compute_state
= fd_bind_compute_state
;
625 pctx
->set_compute_resources
= fd_set_compute_resources
;
626 pctx
->set_global_binding
= fd_set_global_binding
;