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 gallium frontends 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
;
119 fd_resource_set_usage(cb
->buffer
, FD_DIRTY_CONST
);
123 fd_set_shader_buffers(struct pipe_context
*pctx
,
124 enum pipe_shader_type shader
,
125 unsigned start
, unsigned count
,
126 const struct pipe_shader_buffer
*buffers
,
127 unsigned writable_bitmask
)
129 struct fd_context
*ctx
= fd_context(pctx
);
130 struct fd_shaderbuf_stateobj
*so
= &ctx
->shaderbuf
[shader
];
131 const unsigned modified_bits
= u_bit_consecutive(start
, count
);
133 so
->enabled_mask
&= ~modified_bits
;
134 so
->writable_mask
&= ~modified_bits
;
135 so
->writable_mask
|= writable_bitmask
<< start
;
137 for (unsigned i
= 0; i
< count
; i
++) {
138 unsigned n
= i
+ start
;
139 struct pipe_shader_buffer
*buf
= &so
->sb
[n
];
141 if (buffers
&& buffers
[i
].buffer
) {
142 if ((buf
->buffer
== buffers
[i
].buffer
) &&
143 (buf
->buffer_offset
== buffers
[i
].buffer_offset
) &&
144 (buf
->buffer_size
== buffers
[i
].buffer_size
))
147 buf
->buffer_offset
= buffers
[i
].buffer_offset
;
148 buf
->buffer_size
= buffers
[i
].buffer_size
;
149 pipe_resource_reference(&buf
->buffer
, buffers
[i
].buffer
);
151 fd_resource_set_usage(buffers
[i
].buffer
, FD_DIRTY_SSBO
);
153 so
->enabled_mask
|= BIT(n
);
155 pipe_resource_reference(&buf
->buffer
, NULL
);
159 ctx
->dirty_shader
[shader
] |= FD_DIRTY_SHADER_SSBO
;
160 ctx
->dirty
|= FD_DIRTY_SSBO
;
164 fd_set_shader_images(struct pipe_context
*pctx
,
165 enum pipe_shader_type shader
,
166 unsigned start
, unsigned count
,
167 const struct pipe_image_view
*images
)
169 struct fd_context
*ctx
= fd_context(pctx
);
170 struct fd_shaderimg_stateobj
*so
= &ctx
->shaderimg
[shader
];
175 for (unsigned i
= 0; i
< count
; i
++) {
176 unsigned n
= i
+ start
;
177 struct pipe_image_view
*buf
= &so
->si
[n
];
179 if ((buf
->resource
== images
[i
].resource
) &&
180 (buf
->format
== images
[i
].format
) &&
181 (buf
->access
== images
[i
].access
) &&
182 !memcmp(&buf
->u
, &images
[i
].u
, sizeof(buf
->u
)))
186 util_copy_image_view(buf
, &images
[i
]);
189 fd_resource_set_usage(buf
->resource
, FD_DIRTY_IMAGE
);
190 so
->enabled_mask
|= BIT(n
);
192 so
->enabled_mask
&= ~BIT(n
);
196 mask
= (BIT(count
) - 1) << start
;
198 for (unsigned i
= 0; i
< count
; i
++) {
199 unsigned n
= i
+ start
;
200 struct pipe_image_view
*img
= &so
->si
[n
];
202 pipe_resource_reference(&img
->resource
, NULL
);
205 so
->enabled_mask
&= ~mask
;
208 ctx
->dirty_shader
[shader
] |= FD_DIRTY_SHADER_IMAGE
;
209 ctx
->dirty
|= FD_DIRTY_IMAGE
;
213 fd_set_framebuffer_state(struct pipe_context
*pctx
,
214 const struct pipe_framebuffer_state
*framebuffer
)
216 struct fd_context
*ctx
= fd_context(pctx
);
217 struct pipe_framebuffer_state
*cso
;
219 DBG("%ux%u, %u layers, %u samples",
220 framebuffer
->width
, framebuffer
->height
,
221 framebuffer
->layers
, framebuffer
->samples
);
223 cso
= &ctx
->framebuffer
;
225 if (util_framebuffer_state_equal(cso
, framebuffer
))
228 util_copy_framebuffer_state(cso
, framebuffer
);
230 cso
->samples
= util_framebuffer_get_num_samples(cso
);
232 if (ctx
->screen
->reorder
) {
233 struct fd_batch
*old_batch
= NULL
;
235 fd_batch_reference(&old_batch
, ctx
->batch
);
237 if (likely(old_batch
))
238 fd_batch_set_stage(old_batch
, FD_STAGE_NULL
);
240 fd_batch_reference(&ctx
->batch
, NULL
);
241 fd_context_all_dirty(ctx
);
243 if (old_batch
&& old_batch
->blit
&& !old_batch
->back_blit
) {
244 /* for blits, there is not really much point in hanging on
245 * to the uncommitted batch (ie. you probably don't blit
246 * multiple times to the same surface), so we might as
247 * well go ahead and flush this one:
249 fd_batch_flush(old_batch
);
252 fd_batch_reference(&old_batch
, NULL
);
253 } else if (ctx
->batch
) {
254 DBG("%d: cbufs[0]=%p, zsbuf=%p", ctx
->batch
->needs_flush
,
255 framebuffer
->cbufs
[0], framebuffer
->zsbuf
);
256 fd_batch_flush(ctx
->batch
);
259 ctx
->dirty
|= FD_DIRTY_FRAMEBUFFER
;
261 ctx
->disabled_scissor
.minx
= 0;
262 ctx
->disabled_scissor
.miny
= 0;
263 ctx
->disabled_scissor
.maxx
= cso
->width
;
264 ctx
->disabled_scissor
.maxy
= cso
->height
;
266 ctx
->dirty
|= FD_DIRTY_SCISSOR
;
270 fd_set_polygon_stipple(struct pipe_context
*pctx
,
271 const struct pipe_poly_stipple
*stipple
)
273 struct fd_context
*ctx
= fd_context(pctx
);
274 ctx
->stipple
= *stipple
;
275 ctx
->dirty
|= FD_DIRTY_STIPPLE
;
279 fd_set_scissor_states(struct pipe_context
*pctx
,
281 unsigned num_scissors
,
282 const struct pipe_scissor_state
*scissor
)
284 struct fd_context
*ctx
= fd_context(pctx
);
286 ctx
->scissor
= *scissor
;
287 ctx
->dirty
|= FD_DIRTY_SCISSOR
;
291 fd_set_viewport_states(struct pipe_context
*pctx
,
293 unsigned num_viewports
,
294 const struct pipe_viewport_state
*viewport
)
296 struct fd_context
*ctx
= fd_context(pctx
);
297 struct pipe_scissor_state
*scissor
= &ctx
->viewport_scissor
;
298 float minx
, miny
, maxx
, maxy
;
300 ctx
->viewport
= *viewport
;
302 /* see si_get_scissor_from_viewport(): */
304 /* Convert (-1, -1) and (1, 1) from clip space into window space. */
305 minx
= -viewport
->scale
[0] + viewport
->translate
[0];
306 miny
= -viewport
->scale
[1] + viewport
->translate
[1];
307 maxx
= viewport
->scale
[0] + viewport
->translate
[0];
308 maxy
= viewport
->scale
[1] + viewport
->translate
[1];
310 /* Handle inverted viewports. */
318 debug_assert(miny
>= 0);
319 debug_assert(maxy
>= 0);
321 /* Convert to integer and round up the max bounds. */
322 scissor
->minx
= minx
;
323 scissor
->miny
= miny
;
324 scissor
->maxx
= ceilf(maxx
);
325 scissor
->maxy
= ceilf(maxy
);
327 ctx
->dirty
|= FD_DIRTY_VIEWPORT
;
331 fd_set_vertex_buffers(struct pipe_context
*pctx
,
332 unsigned start_slot
, unsigned count
,
333 const struct pipe_vertex_buffer
*vb
)
335 struct fd_context
*ctx
= fd_context(pctx
);
336 struct fd_vertexbuf_stateobj
*so
= &ctx
->vtx
.vertexbuf
;
339 /* on a2xx, pitch is encoded in the vtx fetch instruction, so
340 * we need to mark VTXSTATE as dirty as well to trigger patching
341 * and re-emitting the vtx shader:
343 if (ctx
->screen
->gpu_id
< 300) {
344 for (i
= 0; i
< count
; i
++) {
345 bool new_enabled
= vb
&& vb
[i
].buffer
.resource
;
346 bool old_enabled
= so
->vb
[i
].buffer
.resource
!= NULL
;
347 uint32_t new_stride
= vb
? vb
[i
].stride
: 0;
348 uint32_t old_stride
= so
->vb
[i
].stride
;
349 if ((new_enabled
!= old_enabled
) || (new_stride
!= old_stride
)) {
350 ctx
->dirty
|= FD_DIRTY_VTXSTATE
;
356 util_set_vertex_buffers_mask(so
->vb
, &so
->enabled_mask
, vb
, start_slot
, count
);
357 so
->count
= util_last_bit(so
->enabled_mask
);
362 ctx
->dirty
|= FD_DIRTY_VTXBUF
;
364 for (unsigned i
= 0; i
< count
; i
++) {
365 assert(!vb
[i
].is_user_buffer
);
366 fd_resource_set_usage(vb
[i
].buffer
.resource
, FD_DIRTY_VTXBUF
);
371 fd_blend_state_bind(struct pipe_context
*pctx
, void *hwcso
)
373 struct fd_context
*ctx
= fd_context(pctx
);
374 struct pipe_blend_state
*cso
= hwcso
;
375 bool old_is_dual
= ctx
->blend
?
376 ctx
->blend
->rt
[0].blend_enable
&& util_blend_state_is_dual(ctx
->blend
, 0) :
378 bool new_is_dual
= cso
?
379 cso
->rt
[0].blend_enable
&& util_blend_state_is_dual(cso
, 0) :
382 ctx
->dirty
|= FD_DIRTY_BLEND
;
383 if (old_is_dual
!= new_is_dual
)
384 ctx
->dirty
|= FD_DIRTY_BLEND_DUAL
;
388 fd_blend_state_delete(struct pipe_context
*pctx
, void *hwcso
)
394 fd_rasterizer_state_bind(struct pipe_context
*pctx
, void *hwcso
)
396 struct fd_context
*ctx
= fd_context(pctx
);
397 struct pipe_scissor_state
*old_scissor
= fd_context_get_scissor(ctx
);
398 bool discard
= ctx
->rasterizer
&& ctx
->rasterizer
->rasterizer_discard
;
400 ctx
->rasterizer
= hwcso
;
401 ctx
->dirty
|= FD_DIRTY_RASTERIZER
;
403 if (ctx
->rasterizer
&& ctx
->rasterizer
->scissor
) {
404 ctx
->current_scissor
= &ctx
->scissor
;
406 ctx
->current_scissor
= &ctx
->disabled_scissor
;
409 /* if scissor enable bit changed we need to mark scissor
410 * state as dirty as well:
411 * NOTE: we can do a shallow compare, since we only care
412 * if it changed to/from &ctx->disable_scissor
414 if (old_scissor
!= fd_context_get_scissor(ctx
))
415 ctx
->dirty
|= FD_DIRTY_SCISSOR
;
417 if (ctx
->rasterizer
&& (discard
!= ctx
->rasterizer
->rasterizer_discard
))
418 ctx
->dirty
|= FD_DIRTY_RASTERIZER_DISCARD
;
422 fd_rasterizer_state_delete(struct pipe_context
*pctx
, void *hwcso
)
428 fd_zsa_state_bind(struct pipe_context
*pctx
, void *hwcso
)
430 struct fd_context
*ctx
= fd_context(pctx
);
432 ctx
->dirty
|= FD_DIRTY_ZSA
;
436 fd_zsa_state_delete(struct pipe_context
*pctx
, void *hwcso
)
442 fd_vertex_state_create(struct pipe_context
*pctx
, unsigned num_elements
,
443 const struct pipe_vertex_element
*elements
)
445 struct fd_vertex_stateobj
*so
= CALLOC_STRUCT(fd_vertex_stateobj
);
450 memcpy(so
->pipe
, elements
, sizeof(*elements
) * num_elements
);
451 so
->num_elements
= num_elements
;
457 fd_vertex_state_delete(struct pipe_context
*pctx
, void *hwcso
)
463 fd_vertex_state_bind(struct pipe_context
*pctx
, void *hwcso
)
465 struct fd_context
*ctx
= fd_context(pctx
);
466 ctx
->vtx
.vtx
= hwcso
;
467 ctx
->dirty
|= FD_DIRTY_VTXSTATE
;
470 static struct pipe_stream_output_target
*
471 fd_create_stream_output_target(struct pipe_context
*pctx
,
472 struct pipe_resource
*prsc
, unsigned buffer_offset
,
473 unsigned buffer_size
)
475 struct pipe_stream_output_target
*target
;
476 struct fd_resource
*rsc
= fd_resource(prsc
);
478 target
= CALLOC_STRUCT(pipe_stream_output_target
);
482 pipe_reference_init(&target
->reference
, 1);
483 pipe_resource_reference(&target
->buffer
, prsc
);
485 target
->context
= pctx
;
486 target
->buffer_offset
= buffer_offset
;
487 target
->buffer_size
= buffer_size
;
489 assert(rsc
->base
.target
== PIPE_BUFFER
);
490 util_range_add(&rsc
->base
, &rsc
->valid_buffer_range
,
491 buffer_offset
, buffer_offset
+ buffer_size
);
497 fd_stream_output_target_destroy(struct pipe_context
*pctx
,
498 struct pipe_stream_output_target
*target
)
500 pipe_resource_reference(&target
->buffer
, NULL
);
505 fd_set_stream_output_targets(struct pipe_context
*pctx
,
506 unsigned num_targets
, struct pipe_stream_output_target
**targets
,
507 const unsigned *offsets
)
509 struct fd_context
*ctx
= fd_context(pctx
);
510 struct fd_streamout_stateobj
*so
= &ctx
->streamout
;
513 debug_assert(num_targets
<= ARRAY_SIZE(so
->targets
));
515 for (i
= 0; i
< num_targets
; i
++) {
516 boolean changed
= targets
[i
] != so
->targets
[i
];
517 boolean reset
= (offsets
[i
] != (unsigned)-1);
519 so
->reset
|= (reset
<< i
);
521 if (!changed
&& !reset
)
525 so
->offsets
[i
] = offsets
[i
];
527 pipe_so_target_reference(&so
->targets
[i
], targets
[i
]);
530 for (; i
< so
->num_targets
; i
++) {
531 pipe_so_target_reference(&so
->targets
[i
], NULL
);
534 so
->num_targets
= num_targets
;
536 ctx
->dirty
|= FD_DIRTY_STREAMOUT
;
540 fd_bind_compute_state(struct pipe_context
*pctx
, void *state
)
542 struct fd_context
*ctx
= fd_context(pctx
);
543 ctx
->compute
= state
;
544 ctx
->dirty_shader
[PIPE_SHADER_COMPUTE
] |= FD_DIRTY_SHADER_PROG
;
548 fd_set_compute_resources(struct pipe_context
*pctx
,
549 unsigned start
, unsigned count
, struct pipe_surface
**prscs
)
554 /* used by clover to bind global objects, returning the bo address
558 fd_set_global_binding(struct pipe_context
*pctx
,
559 unsigned first
, unsigned count
, struct pipe_resource
**prscs
,
562 struct fd_context
*ctx
= fd_context(pctx
);
563 struct fd_global_bindings_stateobj
*so
= &ctx
->global_bindings
;
567 for (unsigned i
= 0; i
< count
; i
++) {
568 unsigned n
= i
+ first
;
572 pipe_resource_reference(&so
->buf
[n
], prscs
[i
]);
575 struct fd_resource
*rsc
= fd_resource(so
->buf
[n
]);
576 uint64_t iova
= fd_bo_get_iova(rsc
->bo
);
577 // TODO need to scream if iova > 32b or fix gallium API..
582 so
->enabled_mask
|= BIT(n
);
584 so
->enabled_mask
&= ~BIT(n
);
587 mask
= (BIT(count
) - 1) << first
;
589 for (unsigned i
= 0; i
< count
; i
++) {
590 unsigned n
= i
+ first
;
591 pipe_resource_reference(&so
->buf
[n
], NULL
);
594 so
->enabled_mask
&= ~mask
;
600 fd_state_init(struct pipe_context
*pctx
)
602 pctx
->set_blend_color
= fd_set_blend_color
;
603 pctx
->set_stencil_ref
= fd_set_stencil_ref
;
604 pctx
->set_clip_state
= fd_set_clip_state
;
605 pctx
->set_sample_mask
= fd_set_sample_mask
;
606 pctx
->set_min_samples
= fd_set_min_samples
;
607 pctx
->set_constant_buffer
= fd_set_constant_buffer
;
608 pctx
->set_shader_buffers
= fd_set_shader_buffers
;
609 pctx
->set_shader_images
= fd_set_shader_images
;
610 pctx
->set_framebuffer_state
= fd_set_framebuffer_state
;
611 pctx
->set_polygon_stipple
= fd_set_polygon_stipple
;
612 pctx
->set_scissor_states
= fd_set_scissor_states
;
613 pctx
->set_viewport_states
= fd_set_viewport_states
;
615 pctx
->set_vertex_buffers
= fd_set_vertex_buffers
;
617 pctx
->bind_blend_state
= fd_blend_state_bind
;
618 pctx
->delete_blend_state
= fd_blend_state_delete
;
620 pctx
->bind_rasterizer_state
= fd_rasterizer_state_bind
;
621 pctx
->delete_rasterizer_state
= fd_rasterizer_state_delete
;
623 pctx
->bind_depth_stencil_alpha_state
= fd_zsa_state_bind
;
624 pctx
->delete_depth_stencil_alpha_state
= fd_zsa_state_delete
;
626 if (!pctx
->create_vertex_elements_state
)
627 pctx
->create_vertex_elements_state
= fd_vertex_state_create
;
628 pctx
->delete_vertex_elements_state
= fd_vertex_state_delete
;
629 pctx
->bind_vertex_elements_state
= fd_vertex_state_bind
;
631 pctx
->create_stream_output_target
= fd_create_stream_output_target
;
632 pctx
->stream_output_target_destroy
= fd_stream_output_target_destroy
;
633 pctx
->set_stream_output_targets
= fd_set_stream_output_targets
;
635 if (has_compute(fd_screen(pctx
->screen
))) {
636 pctx
->bind_compute_state
= fd_bind_compute_state
;
637 pctx
->set_compute_resources
= fd_set_compute_resources
;
638 pctx
->set_global_binding
= fd_set_global_binding
;