2 * © Copyright 2018 Alyssa Rosenzweig
3 * Copyright © 2014-2017 Broadcom
4 * Copyright (C) 2017 Intel Corporation
6 * Permission is hereby granted, free of charge, to any person obtaining a
7 * copy of this software and associated documentation files (the "Software"),
8 * to deal in the Software without restriction, including without limitation
9 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10 * and/or sell copies of the Software, and to permit persons to whom the
11 * Software is furnished to do so, subject to the following conditions:
13 * The above copyright notice and this permission notice (including the next
14 * paragraph) shall be included in all copies or substantial portions of the
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31 #include "pan_context.h"
32 #include "pan_minmax_cache.h"
33 #include "panfrost-quirks.h"
35 #include "util/macros.h"
36 #include "util/format/u_format.h"
37 #include "util/u_inlines.h"
38 #include "util/u_upload_mgr.h"
39 #include "util/u_memory.h"
40 #include "util/u_vbuf.h"
41 #include "util/half_float.h"
42 #include "util/u_helpers.h"
43 #include "util/format/u_format.h"
44 #include "util/u_prim.h"
45 #include "util/u_prim_restart.h"
46 #include "indices/u_primconvert.h"
47 #include "tgsi/tgsi_parse.h"
48 #include "tgsi/tgsi_from_mesa.h"
49 #include "util/u_math.h"
51 #include "pan_screen.h"
52 #include "pan_blending.h"
53 #include "pan_blend_shaders.h"
54 #include "pan_cmdstream.h"
56 #include "pandecode/decode.h"
58 struct midgard_tiler_descriptor
59 panfrost_emit_midg_tiler(struct panfrost_batch
*batch
, unsigned vertex_count
)
61 struct panfrost_screen
*screen
= pan_screen(batch
->ctx
->base
.screen
);
62 bool hierarchy
= !(screen
->quirks
& MIDGARD_NO_HIER_TILING
);
63 struct midgard_tiler_descriptor t
= {0};
64 unsigned height
= batch
->key
.height
;
65 unsigned width
= batch
->key
.width
;
68 panfrost_choose_hierarchy_mask(width
, height
, vertex_count
, hierarchy
);
70 /* Compute the polygon header size and use that to offset the body */
72 unsigned header_size
= panfrost_tiler_header_size(
73 width
, height
, t
.hierarchy_mask
, hierarchy
);
75 t
.polygon_list_size
= panfrost_tiler_full_size(
76 width
, height
, t
.hierarchy_mask
, hierarchy
);
81 struct panfrost_bo
*tiler_heap
;
83 tiler_heap
= panfrost_batch_get_tiler_heap(batch
);
84 t
.polygon_list
= panfrost_batch_get_polygon_list(batch
,
89 /* Allow the entire tiler heap */
90 t
.heap_start
= tiler_heap
->gpu
;
91 t
.heap_end
= tiler_heap
->gpu
+ tiler_heap
->size
;
93 struct panfrost_bo
*tiler_dummy
;
95 tiler_dummy
= panfrost_batch_get_tiler_dummy(batch
);
96 header_size
= MALI_TILER_MINIMUM_HEADER_SIZE
;
98 /* The tiler is disabled, so don't allow the tiler heap */
99 t
.heap_start
= tiler_dummy
->gpu
;
100 t
.heap_end
= t
.heap_start
;
102 /* Use a dummy polygon list */
103 t
.polygon_list
= tiler_dummy
->gpu
;
105 /* Disable the tiler */
107 t
.hierarchy_mask
|= MALI_TILER_DISABLED
;
109 t
.hierarchy_mask
= MALI_TILER_USER
;
110 t
.polygon_list_size
= MALI_TILER_MINIMUM_HEADER_SIZE
+ 4;
112 /* We don't have a WRITE_VALUE job, so write the polygon list manually */
113 uint32_t *polygon_list_body
= (uint32_t *) (tiler_dummy
->cpu
+ header_size
);
114 polygon_list_body
[0] = 0xa0000000; /* TODO: Just that? */
118 t
.polygon_list_body
=
119 t
.polygon_list
+ header_size
;
126 struct pipe_context
*pipe
,
128 const union pipe_color_union
*color
,
129 double depth
, unsigned stencil
)
131 struct panfrost_context
*ctx
= pan_context(pipe
);
133 /* TODO: panfrost_get_fresh_batch_for_fbo() instantiates a new batch if
134 * the existing batch targeting this FBO has draws. We could probably
135 * avoid that by replacing plain clears by quad-draws with a specific
136 * color/depth/stencil value, thus avoiding the generation of extra
139 struct panfrost_batch
*batch
= panfrost_get_fresh_batch_for_fbo(ctx
);
141 panfrost_batch_add_fbo_bos(batch
);
142 panfrost_batch_clear(batch
, buffers
, color
, depth
, stencil
);
145 /* Reset per-frame context, called on context initialisation as well as after
146 * flushing a frame */
149 panfrost_invalidate_frame(struct panfrost_context
*ctx
)
151 for (unsigned i
= 0; i
< PIPE_SHADER_TYPES
; ++i
)
152 ctx
->payloads
[i
].postfix
.shared_memory
= 0;
154 /* TODO: When does this need to be handled? */
155 ctx
->active_queries
= true;
158 /* In practice, every field of these payloads should be configurable
159 * arbitrarily, which means these functions are basically catch-all's for
160 * as-of-yet unwavering unknowns */
163 panfrost_emit_vertex_payload(struct panfrost_context
*ctx
)
165 /* 0x2 bit clear on 32-bit T6XX */
167 struct midgard_payload_vertex_tiler payload
= {
168 .gl_enables
= 0x4 | 0x2,
171 /* Vertex and compute are closely coupled, so share a payload */
173 memcpy(&ctx
->payloads
[PIPE_SHADER_VERTEX
], &payload
, sizeof(payload
));
174 memcpy(&ctx
->payloads
[PIPE_SHADER_COMPUTE
], &payload
, sizeof(payload
));
178 panfrost_writes_point_size(struct panfrost_context
*ctx
)
180 assert(ctx
->shader
[PIPE_SHADER_VERTEX
]);
181 struct panfrost_shader_state
*vs
= panfrost_get_shader_state(ctx
, PIPE_SHADER_VERTEX
);
183 return vs
->writes_point_size
&& ctx
->payloads
[PIPE_SHADER_FRAGMENT
].prefix
.draw_mode
== MALI_POINTS
;
186 /* Stage the attribute descriptors so we can adjust src_offset
187 * to let BOs align nicely */
190 panfrost_stage_attributes(struct panfrost_context
*ctx
)
192 struct panfrost_batch
*batch
= panfrost_get_batch_for_fbo(ctx
);
193 struct panfrost_vertex_state
*so
= ctx
->vertex
;
195 size_t sz
= sizeof(struct mali_attr_meta
) * PAN_MAX_ATTRIBUTE
;
196 struct panfrost_transfer transfer
= panfrost_allocate_transient(batch
, sz
);
197 struct mali_attr_meta
*target
= (struct mali_attr_meta
*) transfer
.cpu
;
199 /* Copy as-is for the first pass */
200 memcpy(target
, so
->hw
, sz
);
202 /* Fixup offsets for the second pass. Recall that the hardware
203 * calculates attribute addresses as:
205 * addr = base + (stride * vtx) + src_offset;
207 * However, on Mali, base must be aligned to 64-bytes, so we
210 * base' = base & ~63 = base - (base & 63)
212 * To compensate when using base' (see emit_vertex_data), we have
213 * to adjust src_offset by the masked off piece:
215 * addr' = base' + (stride * vtx) + (src_offset + (base & 63))
216 * = base - (base & 63) + (stride * vtx) + src_offset + (base & 63)
217 * = base + (stride * vtx) + src_offset
223 unsigned start
= ctx
->payloads
[PIPE_SHADER_VERTEX
].offset_start
;
225 for (unsigned i
= 0; i
< so
->num_elements
; ++i
) {
226 unsigned vbi
= so
->pipe
[i
].vertex_buffer_index
;
227 struct pipe_vertex_buffer
*buf
= &ctx
->vertex_buffers
[vbi
];
228 struct panfrost_resource
*rsrc
= (struct panfrost_resource
*) (buf
->buffer
.resource
);
229 mali_ptr addr
= rsrc
->bo
->gpu
+ buf
->buffer_offset
;
231 /* Adjust by the masked off bits of the offset. Make sure we
232 * read src_offset from so->hw (which is not GPU visible)
233 * rather than target (which is) due to caching effects */
235 unsigned src_offset
= so
->hw
[i
].src_offset
;
236 src_offset
+= (addr
& 63);
238 /* Also, somewhat obscurely per-instance data needs to be
239 * offset in response to a delayed start in an indexed draw */
241 if (so
->pipe
[i
].instance_divisor
&& ctx
->instance_count
> 1 && start
)
242 src_offset
-= buf
->stride
* start
;
244 target
[i
].src_offset
= src_offset
;
247 /* Let's also include vertex builtins */
249 struct mali_attr_meta builtin
= {
250 .format
= MALI_R32UI
,
251 .swizzle
= panfrost_get_default_swizzle(1)
254 /* See mali_attr_meta specification for the magic number */
256 builtin
.index
= so
->vertexid_index
;
257 memcpy(&target
[PAN_VERTEX_ID
], &builtin
, 4);
259 builtin
.index
= so
->vertexid_index
+ 1;
260 memcpy(&target
[PAN_INSTANCE_ID
], &builtin
, 4);
262 ctx
->payloads
[PIPE_SHADER_VERTEX
].postfix
.attribute_meta
= transfer
.gpu
;
265 /* Compute number of UBOs active (more specifically, compute the highest UBO
266 * number addressable -- if there are gaps, include them in the count anyway).
267 * We always include UBO #0 in the count, since we *need* uniforms enabled for
271 panfrost_ubo_count(struct panfrost_context
*ctx
, enum pipe_shader_type stage
)
273 unsigned mask
= ctx
->constant_buffer
[stage
].enabled_mask
| 1;
274 return 32 - __builtin_clz(mask
);
277 /* Go through dirty flags and actualise them in the cmdstream. */
280 panfrost_emit_for_draw(struct panfrost_context
*ctx
)
282 struct panfrost_batch
*batch
= panfrost_get_batch_for_fbo(ctx
);
284 panfrost_batch_add_fbo_bos(batch
);
286 for (int i
= 0; i
<= PIPE_SHADER_FRAGMENT
; ++i
)
287 panfrost_vt_attach_framebuffer(ctx
, &ctx
->payloads
[i
]);
289 panfrost_emit_vertex_data(batch
);
291 /* Varyings emitted for -all- geometry */
292 unsigned total_count
= ctx
->padded_count
* ctx
->instance_count
;
293 panfrost_emit_varying_descriptor(ctx
, total_count
);
295 panfrost_batch_set_requirements(batch
);
297 panfrost_vt_update_rasterizer(ctx
, &ctx
->payloads
[PIPE_SHADER_FRAGMENT
]);
298 panfrost_vt_update_occlusion_query(ctx
, &ctx
->payloads
[PIPE_SHADER_FRAGMENT
]);
300 panfrost_emit_shader_meta(batch
, PIPE_SHADER_VERTEX
,
301 &ctx
->payloads
[PIPE_SHADER_VERTEX
]);
302 panfrost_emit_shader_meta(batch
, PIPE_SHADER_FRAGMENT
,
303 &ctx
->payloads
[PIPE_SHADER_FRAGMENT
]);
305 /* We stage to transient, so always dirty.. */
307 panfrost_stage_attributes(ctx
);
309 for (int i
= 0; i
<= PIPE_SHADER_FRAGMENT
; ++i
) {
310 panfrost_emit_sampler_descriptors(batch
, i
, &ctx
->payloads
[i
]);
311 panfrost_emit_texture_descriptors(batch
, i
, &ctx
->payloads
[i
]);
312 panfrost_emit_const_buf(batch
, i
, &ctx
->payloads
[i
]);
315 /* TODO: Upload the viewport somewhere more appropriate */
317 panfrost_emit_viewport(batch
, &ctx
->payloads
[PIPE_SHADER_FRAGMENT
]);
320 /* Corresponds to exactly one draw, but does not submit anything */
323 panfrost_queue_draw(struct panfrost_context
*ctx
)
325 /* Handle dirty flags now */
326 panfrost_emit_for_draw(ctx
);
328 /* If rasterizer discard is enable, only submit the vertex */
330 bool rasterizer_discard
= ctx
->rasterizer
331 && ctx
->rasterizer
->base
.rasterizer_discard
;
334 struct midgard_payload_vertex_tiler
*vertex_payload
= &ctx
->payloads
[PIPE_SHADER_VERTEX
];
335 struct midgard_payload_vertex_tiler
*tiler_payload
= &ctx
->payloads
[PIPE_SHADER_FRAGMENT
];
337 struct panfrost_batch
*batch
= panfrost_get_batch_for_fbo(ctx
);
338 bool wallpapering
= ctx
->wallpaper_batch
&& batch
->tiler_dep
;
341 /* Inject in reverse order, with "predicted" job indices. THIS IS A HACK XXX */
342 panfrost_new_job(batch
, JOB_TYPE_TILER
, false, batch
->job_index
+ 2, tiler_payload
, sizeof(*tiler_payload
), true);
343 panfrost_new_job(batch
, JOB_TYPE_VERTEX
, false, 0, vertex_payload
, sizeof(*vertex_payload
), true);
345 unsigned vertex
= panfrost_new_job(batch
, JOB_TYPE_VERTEX
, false, 0, vertex_payload
, sizeof(*vertex_payload
), false);
347 if (!rasterizer_discard
)
348 panfrost_new_job(batch
, JOB_TYPE_TILER
, false, vertex
, tiler_payload
, sizeof(*tiler_payload
), false);
351 panfrost_batch_adjust_stack_size(batch
);
354 /* The entire frame is in memory -- send it off to the kernel! */
358 struct pipe_context
*pipe
,
359 struct pipe_fence_handle
**fence
,
362 struct panfrost_context
*ctx
= pan_context(pipe
);
363 struct util_dynarray fences
;
365 /* We must collect the fences before the flush is done, otherwise we'll
366 * lose track of them.
369 util_dynarray_init(&fences
, NULL
);
370 hash_table_foreach(ctx
->batches
, hentry
) {
371 struct panfrost_batch
*batch
= hentry
->data
;
373 panfrost_batch_fence_reference(batch
->out_sync
);
374 util_dynarray_append(&fences
,
375 struct panfrost_batch_fence
*,
380 /* Submit all pending jobs */
381 panfrost_flush_all_batches(ctx
, false);
384 struct panfrost_fence
*f
= panfrost_fence_create(ctx
, &fences
);
385 pipe
->screen
->fence_reference(pipe
->screen
, fence
, NULL
);
386 *fence
= (struct pipe_fence_handle
*)f
;
388 util_dynarray_foreach(&fences
, struct panfrost_batch_fence
*, fence
)
389 panfrost_batch_fence_unreference(*fence
);
391 util_dynarray_fini(&fences
);
394 if (pan_debug
& PAN_DBG_TRACE
)
395 pandecode_next_frame();
398 #define DEFINE_CASE(c) case PIPE_PRIM_##c: return MALI_##c;
401 g2m_draw_mode(enum pipe_prim_type mode
)
406 DEFINE_CASE(LINE_LOOP
);
407 DEFINE_CASE(LINE_STRIP
);
408 DEFINE_CASE(TRIANGLES
);
409 DEFINE_CASE(TRIANGLE_STRIP
);
410 DEFINE_CASE(TRIANGLE_FAN
);
412 DEFINE_CASE(QUAD_STRIP
);
413 DEFINE_CASE(POLYGON
);
416 unreachable("Invalid draw mode");
423 panfrost_translate_index_size(unsigned size
)
427 return MALI_DRAW_INDEXED_UINT8
;
430 return MALI_DRAW_INDEXED_UINT16
;
433 return MALI_DRAW_INDEXED_UINT32
;
436 unreachable("Invalid index size");
440 /* Gets a GPU address for the associated index buffer. Only gauranteed to be
441 * good for the duration of the draw (transient), could last longer. Also get
442 * the bounds on the index buffer for the range accessed by the draw. We do
443 * these operations together because there are natural optimizations which
444 * require them to be together. */
447 panfrost_get_index_buffer_bounded(struct panfrost_context
*ctx
, const struct pipe_draw_info
*info
, unsigned *min_index
, unsigned *max_index
)
449 struct panfrost_resource
*rsrc
= (struct panfrost_resource
*) (info
->index
.resource
);
451 off_t offset
= info
->start
* info
->index_size
;
452 struct panfrost_batch
*batch
= panfrost_get_batch_for_fbo(ctx
);
455 bool needs_indices
= true;
457 if (info
->max_index
!= ~0u) {
458 *min_index
= info
->min_index
;
459 *max_index
= info
->max_index
;
460 needs_indices
= false;
463 if (!info
->has_user_indices
) {
464 /* Only resources can be directly mapped */
465 panfrost_batch_add_bo(batch
, rsrc
->bo
,
466 PAN_BO_ACCESS_SHARED
|
468 PAN_BO_ACCESS_VERTEX_TILER
);
469 out
= rsrc
->bo
->gpu
+ offset
;
471 /* Check the cache */
472 needs_indices
= !panfrost_minmax_cache_get(rsrc
->index_cache
, info
->start
, info
->count
,
473 min_index
, max_index
);
475 /* Otherwise, we need to upload to transient memory */
476 const uint8_t *ibuf8
= (const uint8_t *) info
->index
.user
;
477 out
= panfrost_upload_transient(batch
, ibuf8
+ offset
, info
->count
* info
->index_size
);
482 u_vbuf_get_minmax_index(&ctx
->base
, info
, min_index
, max_index
);
484 if (!info
->has_user_indices
) {
485 panfrost_minmax_cache_add(rsrc
->index_cache
, info
->start
, info
->count
,
486 *min_index
, *max_index
);
495 panfrost_scissor_culls_everything(struct panfrost_context
*ctx
)
497 const struct pipe_scissor_state
*ss
= &ctx
->scissor
;
499 /* Check if we're scissoring at all */
501 if (!(ctx
->rasterizer
&& ctx
->rasterizer
->base
.scissor
))
504 return (ss
->minx
== ss
->maxx
) || (ss
->miny
== ss
->maxy
);
507 /* Count generated primitives (when there is no geom/tess shaders) for
508 * transform feedback */
511 panfrost_statistics_record(
512 struct panfrost_context
*ctx
,
513 const struct pipe_draw_info
*info
)
515 if (!ctx
->active_queries
)
518 uint32_t prims
= u_prims_for_vertices(info
->mode
, info
->count
);
519 ctx
->prims_generated
+= prims
;
521 if (!ctx
->streamout
.num_targets
)
524 ctx
->tf_prims_generated
+= prims
;
529 struct pipe_context
*pipe
,
530 const struct pipe_draw_info
*info
)
532 struct panfrost_context
*ctx
= pan_context(pipe
);
534 /* First of all, check the scissor to see if anything is drawn at all.
535 * If it's not, we drop the draw (mostly a conformance issue;
536 * well-behaved apps shouldn't hit this) */
538 if (panfrost_scissor_culls_everything(ctx
))
541 int mode
= info
->mode
;
543 /* Fallback unsupported restart index */
544 unsigned primitive_index
= (1 << (info
->index_size
* 8)) - 1;
546 if (info
->primitive_restart
&& info
->index_size
547 && info
->restart_index
!= primitive_index
) {
548 util_draw_vbo_without_prim_restart(pipe
, info
);
552 /* Fallback for unsupported modes */
554 assert(ctx
->rasterizer
!= NULL
);
556 if (!(ctx
->draw_modes
& (1 << mode
))) {
557 if (mode
== PIPE_PRIM_QUADS
&& info
->count
== 4 && !ctx
->rasterizer
->base
.flatshade
) {
558 mode
= PIPE_PRIM_TRIANGLE_FAN
;
560 if (info
->count
< 4) {
561 /* Degenerate case? */
565 util_primconvert_save_rasterizer_state(ctx
->primconvert
, &ctx
->rasterizer
->base
);
566 util_primconvert_draw_vbo(ctx
->primconvert
, info
);
571 ctx
->payloads
[PIPE_SHADER_VERTEX
].offset_start
= info
->start
;
572 ctx
->payloads
[PIPE_SHADER_FRAGMENT
].offset_start
= info
->start
;
574 /* Now that we have a guaranteed terminating path, find the job.
575 * Assignment commented out to prevent unused warning */
577 /* struct panfrost_batch *batch = */ panfrost_get_batch_for_fbo(ctx
);
579 ctx
->payloads
[PIPE_SHADER_FRAGMENT
].prefix
.draw_mode
= g2m_draw_mode(mode
);
581 /* Take into account a negative bias */
582 ctx
->vertex_count
= info
->count
+ abs(info
->index_bias
);
583 ctx
->instance_count
= info
->instance_count
;
584 ctx
->active_prim
= info
->mode
;
586 /* For non-indexed draws, they're the same */
587 unsigned vertex_count
= ctx
->vertex_count
;
589 unsigned draw_flags
= 0;
591 /* The draw flags interpret how primitive size is interpreted */
593 if (panfrost_writes_point_size(ctx
))
594 draw_flags
|= MALI_DRAW_VARYING_SIZE
;
596 if (info
->primitive_restart
)
597 draw_flags
|= MALI_DRAW_PRIMITIVE_RESTART_FIXED_INDEX
;
599 /* These doesn't make much sense */
601 draw_flags
|= 0x3000;
603 if (ctx
->rasterizer
&& ctx
->rasterizer
->base
.flatshade_first
)
604 draw_flags
|= MALI_DRAW_FLATSHADE_FIRST
;
606 panfrost_statistics_record(ctx
, info
);
608 if (info
->index_size
) {
609 unsigned min_index
= 0, max_index
= 0;
610 ctx
->payloads
[PIPE_SHADER_FRAGMENT
].prefix
.indices
=
611 panfrost_get_index_buffer_bounded(ctx
, info
, &min_index
, &max_index
);
613 /* Use the corresponding values */
614 vertex_count
= max_index
- min_index
+ 1;
615 ctx
->payloads
[PIPE_SHADER_VERTEX
].offset_start
= min_index
+ info
->index_bias
;
616 ctx
->payloads
[PIPE_SHADER_FRAGMENT
].offset_start
= min_index
+ info
->index_bias
;
618 ctx
->payloads
[PIPE_SHADER_FRAGMENT
].prefix
.offset_bias_correction
= -min_index
;
619 ctx
->payloads
[PIPE_SHADER_FRAGMENT
].prefix
.index_count
= MALI_POSITIVE(info
->count
);
621 draw_flags
|= panfrost_translate_index_size(info
->index_size
);
623 /* Index count == vertex count, if no indexing is applied, as
624 * if it is internally indexed in the expected order */
626 ctx
->payloads
[PIPE_SHADER_FRAGMENT
].prefix
.offset_bias_correction
= 0;
627 ctx
->payloads
[PIPE_SHADER_FRAGMENT
].prefix
.index_count
= MALI_POSITIVE(ctx
->vertex_count
);
629 /* Reverse index state */
630 ctx
->payloads
[PIPE_SHADER_FRAGMENT
].prefix
.indices
= (mali_ptr
) 0;
633 /* Dispatch "compute jobs" for the vertex/tiler pair as (1,
634 * vertex_count, 1) */
636 panfrost_pack_work_groups_fused(
637 &ctx
->payloads
[PIPE_SHADER_VERTEX
].prefix
,
638 &ctx
->payloads
[PIPE_SHADER_FRAGMENT
].prefix
,
639 1, vertex_count
, info
->instance_count
,
642 ctx
->payloads
[PIPE_SHADER_FRAGMENT
].prefix
.unknown_draw
= draw_flags
;
644 /* Encode the padded vertex count */
646 if (info
->instance_count
> 1) {
647 ctx
->padded_count
= panfrost_padded_vertex_count(vertex_count
);
649 unsigned shift
= __builtin_ctz(ctx
->padded_count
);
650 unsigned k
= ctx
->padded_count
>> (shift
+ 1);
652 ctx
->payloads
[PIPE_SHADER_VERTEX
].instance_shift
= shift
;
653 ctx
->payloads
[PIPE_SHADER_FRAGMENT
].instance_shift
= shift
;
655 ctx
->payloads
[PIPE_SHADER_VERTEX
].instance_odd
= k
;
656 ctx
->payloads
[PIPE_SHADER_FRAGMENT
].instance_odd
= k
;
658 ctx
->padded_count
= vertex_count
;
660 /* Reset instancing state */
661 ctx
->payloads
[PIPE_SHADER_VERTEX
].instance_shift
= 0;
662 ctx
->payloads
[PIPE_SHADER_VERTEX
].instance_odd
= 0;
663 ctx
->payloads
[PIPE_SHADER_FRAGMENT
].instance_shift
= 0;
664 ctx
->payloads
[PIPE_SHADER_FRAGMENT
].instance_odd
= 0;
667 /* Fire off the draw itself */
668 panfrost_queue_draw(ctx
);
670 /* Increment transform feedback offsets */
672 for (unsigned i
= 0; i
< ctx
->streamout
.num_targets
; ++i
) {
673 unsigned output_count
= u_stream_outputs_for_vertices(
674 ctx
->active_prim
, ctx
->vertex_count
);
676 ctx
->streamout
.offsets
[i
] += output_count
;
683 panfrost_generic_cso_delete(struct pipe_context
*pctx
, void *hwcso
)
689 panfrost_create_rasterizer_state(
690 struct pipe_context
*pctx
,
691 const struct pipe_rasterizer_state
*cso
)
693 struct panfrost_rasterizer
*so
= CALLOC_STRUCT(panfrost_rasterizer
);
701 panfrost_bind_rasterizer_state(
702 struct pipe_context
*pctx
,
705 struct panfrost_context
*ctx
= pan_context(pctx
);
707 ctx
->rasterizer
= hwcso
;
712 /* Gauranteed with the core GL call, so don't expose ARB_polygon_offset */
713 assert(ctx
->rasterizer
->base
.offset_clamp
== 0.0);
715 /* Point sprites are emulated */
717 struct panfrost_shader_state
*variant
= panfrost_get_shader_state(ctx
, PIPE_SHADER_FRAGMENT
);
719 if (ctx
->rasterizer
->base
.sprite_coord_enable
|| (variant
&& variant
->point_sprite_mask
))
720 ctx
->base
.bind_fs_state(&ctx
->base
, ctx
->shader
[PIPE_SHADER_FRAGMENT
]);
724 panfrost_create_vertex_elements_state(
725 struct pipe_context
*pctx
,
726 unsigned num_elements
,
727 const struct pipe_vertex_element
*elements
)
729 struct panfrost_vertex_state
*so
= CALLOC_STRUCT(panfrost_vertex_state
);
731 so
->num_elements
= num_elements
;
732 memcpy(so
->pipe
, elements
, sizeof(*elements
) * num_elements
);
734 for (int i
= 0; i
< num_elements
; ++i
) {
737 enum pipe_format fmt
= elements
[i
].src_format
;
738 const struct util_format_description
*desc
= util_format_description(fmt
);
739 so
->hw
[i
].unknown1
= 0x2;
740 so
->hw
[i
].swizzle
= panfrost_get_default_swizzle(desc
->nr_channels
);
742 so
->hw
[i
].format
= panfrost_find_format(desc
);
744 /* The field itself should probably be shifted over */
745 so
->hw
[i
].src_offset
= elements
[i
].src_offset
;
752 panfrost_bind_vertex_elements_state(
753 struct pipe_context
*pctx
,
756 struct panfrost_context
*ctx
= pan_context(pctx
);
761 panfrost_create_shader_state(
762 struct pipe_context
*pctx
,
763 const struct pipe_shader_state
*cso
,
764 enum pipe_shader_type stage
)
766 struct panfrost_shader_variants
*so
= CALLOC_STRUCT(panfrost_shader_variants
);
769 /* Token deep copy to prevent memory corruption */
771 if (cso
->type
== PIPE_SHADER_IR_TGSI
)
772 so
->base
.tokens
= tgsi_dup_tokens(so
->base
.tokens
);
774 /* Precompile for shader-db if we need to */
775 if (unlikely((pan_debug
& PAN_DBG_PRECOMPILE
) && cso
->type
== PIPE_SHADER_IR_NIR
)) {
776 struct panfrost_context
*ctx
= pan_context(pctx
);
778 struct panfrost_shader_state state
;
779 uint64_t outputs_written
;
781 panfrost_shader_compile(ctx
, PIPE_SHADER_IR_NIR
,
783 tgsi_processor_to_shader_stage(stage
),
784 &state
, &outputs_written
);
791 panfrost_delete_shader_state(
792 struct pipe_context
*pctx
,
795 struct panfrost_shader_variants
*cso
= (struct panfrost_shader_variants
*) so
;
797 if (cso
->base
.type
== PIPE_SHADER_IR_TGSI
) {
798 DBG("Deleting TGSI shader leaks duplicated tokens\n");
801 for (unsigned i
= 0; i
< cso
->variant_count
; ++i
) {
802 struct panfrost_shader_state
*shader_state
= &cso
->variants
[i
];
803 panfrost_bo_unreference(shader_state
->bo
);
804 shader_state
->bo
= NULL
;
812 panfrost_create_sampler_state(
813 struct pipe_context
*pctx
,
814 const struct pipe_sampler_state
*cso
)
816 struct panfrost_sampler_state
*so
= CALLOC_STRUCT(panfrost_sampler_state
);
819 panfrost_sampler_desc_init(cso
, &so
->hw
);
825 panfrost_bind_sampler_states(
826 struct pipe_context
*pctx
,
827 enum pipe_shader_type shader
,
828 unsigned start_slot
, unsigned num_sampler
,
831 assert(start_slot
== 0);
833 struct panfrost_context
*ctx
= pan_context(pctx
);
835 /* XXX: Should upload, not just copy? */
836 ctx
->sampler_count
[shader
] = num_sampler
;
837 memcpy(ctx
->samplers
[shader
], sampler
, num_sampler
* sizeof (void *));
841 panfrost_variant_matches(
842 struct panfrost_context
*ctx
,
843 struct panfrost_shader_state
*variant
,
844 enum pipe_shader_type type
)
846 struct pipe_rasterizer_state
*rasterizer
= &ctx
->rasterizer
->base
;
847 struct pipe_alpha_state
*alpha
= &ctx
->depth_stencil
->alpha
;
849 bool is_fragment
= (type
== PIPE_SHADER_FRAGMENT
);
851 if (is_fragment
&& (alpha
->enabled
|| variant
->alpha_state
.enabled
)) {
852 /* Make sure enable state is at least the same */
853 if (alpha
->enabled
!= variant
->alpha_state
.enabled
) {
857 /* Check that the contents of the test are the same */
858 bool same_func
= alpha
->func
== variant
->alpha_state
.func
;
859 bool same_ref
= alpha
->ref_value
== variant
->alpha_state
.ref_value
;
861 if (!(same_func
&& same_ref
)) {
866 if (is_fragment
&& rasterizer
&& (rasterizer
->sprite_coord_enable
|
867 variant
->point_sprite_mask
)) {
868 /* Ensure the same varyings are turned to point sprites */
869 if (rasterizer
->sprite_coord_enable
!= variant
->point_sprite_mask
)
872 /* Ensure the orientation is correct */
874 rasterizer
->sprite_coord_mode
==
875 PIPE_SPRITE_COORD_UPPER_LEFT
;
877 if (variant
->point_sprite_upper_left
!= upper_left
)
881 /* Otherwise, we're good to go */
886 * Fix an uncompiled shader's stream output info, and produce a bitmask
887 * of which VARYING_SLOT_* are captured for stream output.
889 * Core Gallium stores output->register_index as a "slot" number, where
890 * slots are assigned consecutively to all outputs in info->outputs_written.
891 * This naive packing of outputs doesn't work for us - we too have slots,
892 * but the layout is defined by the VUE map, which we won't have until we
893 * compile a specific shader variant. So, we remap these and simply store
894 * VARYING_SLOT_* in our copy's output->register_index fields.
896 * We then produce a bitmask of outputs which are used for SO.
898 * Implementation from iris.
902 update_so_info(struct pipe_stream_output_info
*so_info
,
903 uint64_t outputs_written
)
905 uint64_t so_outputs
= 0;
906 uint8_t reverse_map
[64] = {0};
909 while (outputs_written
)
910 reverse_map
[slot
++] = u_bit_scan64(&outputs_written
);
912 for (unsigned i
= 0; i
< so_info
->num_outputs
; i
++) {
913 struct pipe_stream_output
*output
= &so_info
->output
[i
];
915 /* Map Gallium's condensed "slots" back to real VARYING_SLOT_* enums */
916 output
->register_index
= reverse_map
[output
->register_index
];
918 so_outputs
|= 1ull << output
->register_index
;
925 panfrost_bind_shader_state(
926 struct pipe_context
*pctx
,
928 enum pipe_shader_type type
)
930 struct panfrost_context
*ctx
= pan_context(pctx
);
931 ctx
->shader
[type
] = hwcso
;
935 /* Match the appropriate variant */
938 struct panfrost_shader_variants
*variants
= (struct panfrost_shader_variants
*) hwcso
;
940 for (unsigned i
= 0; i
< variants
->variant_count
; ++i
) {
941 if (panfrost_variant_matches(ctx
, &variants
->variants
[i
], type
)) {
948 /* No variant matched, so create a new one */
949 variant
= variants
->variant_count
++;
951 if (variants
->variant_count
> variants
->variant_space
) {
952 unsigned old_space
= variants
->variant_space
;
954 variants
->variant_space
*= 2;
955 if (variants
->variant_space
== 0)
956 variants
->variant_space
= 1;
958 /* Arbitrary limit to stop runaway programs from
959 * creating an unbounded number of shader variants. */
960 assert(variants
->variant_space
< 1024);
962 unsigned msize
= sizeof(struct panfrost_shader_state
);
963 variants
->variants
= realloc(variants
->variants
,
964 variants
->variant_space
* msize
);
966 memset(&variants
->variants
[old_space
], 0,
967 (variants
->variant_space
- old_space
) * msize
);
970 struct panfrost_shader_state
*v
=
971 &variants
->variants
[variant
];
973 if (type
== PIPE_SHADER_FRAGMENT
) {
974 v
->alpha_state
= ctx
->depth_stencil
->alpha
;
976 if (ctx
->rasterizer
) {
977 v
->point_sprite_mask
= ctx
->rasterizer
->base
.sprite_coord_enable
;
978 v
->point_sprite_upper_left
=
979 ctx
->rasterizer
->base
.sprite_coord_mode
==
980 PIPE_SPRITE_COORD_UPPER_LEFT
;
985 /* Select this variant */
986 variants
->active_variant
= variant
;
988 struct panfrost_shader_state
*shader_state
= &variants
->variants
[variant
];
989 assert(panfrost_variant_matches(ctx
, shader_state
, type
));
991 /* We finally have a variant, so compile it */
993 if (!shader_state
->compiled
) {
994 uint64_t outputs_written
= 0;
996 panfrost_shader_compile(ctx
, variants
->base
.type
,
997 variants
->base
.type
== PIPE_SHADER_IR_NIR
?
998 variants
->base
.ir
.nir
:
999 variants
->base
.tokens
,
1000 tgsi_processor_to_shader_stage(type
),
1004 shader_state
->compiled
= true;
1006 /* Fixup the stream out information, since what Gallium returns
1007 * normally is mildly insane */
1009 shader_state
->stream_output
= variants
->base
.stream_output
;
1010 shader_state
->so_mask
=
1011 update_so_info(&shader_state
->stream_output
, outputs_written
);
1016 panfrost_create_vs_state(struct pipe_context
*pctx
, const struct pipe_shader_state
*hwcso
)
1018 return panfrost_create_shader_state(pctx
, hwcso
, PIPE_SHADER_VERTEX
);
1022 panfrost_create_fs_state(struct pipe_context
*pctx
, const struct pipe_shader_state
*hwcso
)
1024 return panfrost_create_shader_state(pctx
, hwcso
, PIPE_SHADER_FRAGMENT
);
1028 panfrost_bind_vs_state(struct pipe_context
*pctx
, void *hwcso
)
1030 panfrost_bind_shader_state(pctx
, hwcso
, PIPE_SHADER_VERTEX
);
1034 panfrost_bind_fs_state(struct pipe_context
*pctx
, void *hwcso
)
1036 panfrost_bind_shader_state(pctx
, hwcso
, PIPE_SHADER_FRAGMENT
);
1040 panfrost_set_vertex_buffers(
1041 struct pipe_context
*pctx
,
1042 unsigned start_slot
,
1043 unsigned num_buffers
,
1044 const struct pipe_vertex_buffer
*buffers
)
1046 struct panfrost_context
*ctx
= pan_context(pctx
);
1048 util_set_vertex_buffers_mask(ctx
->vertex_buffers
, &ctx
->vb_mask
, buffers
, start_slot
, num_buffers
);
1052 panfrost_set_constant_buffer(
1053 struct pipe_context
*pctx
,
1054 enum pipe_shader_type shader
, uint index
,
1055 const struct pipe_constant_buffer
*buf
)
1057 struct panfrost_context
*ctx
= pan_context(pctx
);
1058 struct panfrost_constant_buffer
*pbuf
= &ctx
->constant_buffer
[shader
];
1060 util_copy_constant_buffer(&pbuf
->cb
[index
], buf
);
1062 unsigned mask
= (1 << index
);
1064 if (unlikely(!buf
)) {
1065 pbuf
->enabled_mask
&= ~mask
;
1066 pbuf
->dirty_mask
&= ~mask
;
1070 pbuf
->enabled_mask
|= mask
;
1071 pbuf
->dirty_mask
|= mask
;
1075 panfrost_set_stencil_ref(
1076 struct pipe_context
*pctx
,
1077 const struct pipe_stencil_ref
*ref
)
1079 struct panfrost_context
*ctx
= pan_context(pctx
);
1080 ctx
->stencil_ref
= *ref
;
1083 static enum mali_texture_type
1084 panfrost_translate_texture_type(enum pipe_texture_target t
) {
1088 case PIPE_TEXTURE_1D
:
1089 case PIPE_TEXTURE_1D_ARRAY
:
1092 case PIPE_TEXTURE_2D
:
1093 case PIPE_TEXTURE_2D_ARRAY
:
1094 case PIPE_TEXTURE_RECT
:
1097 case PIPE_TEXTURE_3D
:
1100 case PIPE_TEXTURE_CUBE
:
1101 case PIPE_TEXTURE_CUBE_ARRAY
:
1102 return MALI_TEX_CUBE
;
1105 unreachable("Unknown target");
1109 static struct pipe_sampler_view
*
1110 panfrost_create_sampler_view(
1111 struct pipe_context
*pctx
,
1112 struct pipe_resource
*texture
,
1113 const struct pipe_sampler_view
*template)
1115 struct panfrost_screen
*screen
= pan_screen(pctx
->screen
);
1116 struct panfrost_sampler_view
*so
= rzalloc(pctx
, struct panfrost_sampler_view
);
1118 pipe_reference(NULL
, &texture
->reference
);
1120 struct panfrost_resource
*prsrc
= (struct panfrost_resource
*) texture
;
1123 so
->base
= *template;
1124 so
->base
.texture
= texture
;
1125 so
->base
.reference
.count
= 1;
1126 so
->base
.context
= pctx
;
1128 unsigned char user_swizzle
[4] = {
1129 template->swizzle_r
,
1130 template->swizzle_g
,
1131 template->swizzle_b
,
1135 /* In the hardware, array_size refers specifically to array textures,
1136 * whereas in Gallium, it also covers cubemaps */
1138 unsigned array_size
= texture
->array_size
;
1140 if (template->target
== PIPE_TEXTURE_CUBE
) {
1141 /* TODO: Cubemap arrays */
1142 assert(array_size
== 6);
1146 enum mali_texture_type type
=
1147 panfrost_translate_texture_type(template->target
);
1149 unsigned size
= panfrost_estimate_texture_size(
1150 template->u
.tex
.first_level
,
1151 template->u
.tex
.last_level
,
1152 template->u
.tex
.first_layer
,
1153 template->u
.tex
.last_layer
,
1154 type
, prsrc
->layout
);
1156 so
->bo
= panfrost_bo_create(screen
, size
, 0);
1158 panfrost_new_texture(
1160 texture
->width0
, texture
->height0
,
1161 texture
->depth0
, array_size
,
1163 type
, prsrc
->layout
,
1164 template->u
.tex
.first_level
,
1165 template->u
.tex
.last_level
,
1166 template->u
.tex
.first_layer
,
1167 template->u
.tex
.last_layer
,
1168 prsrc
->cubemap_stride
,
1169 panfrost_translate_swizzle_4(user_swizzle
),
1173 return (struct pipe_sampler_view
*) so
;
1177 panfrost_set_sampler_views(
1178 struct pipe_context
*pctx
,
1179 enum pipe_shader_type shader
,
1180 unsigned start_slot
, unsigned num_views
,
1181 struct pipe_sampler_view
**views
)
1183 struct panfrost_context
*ctx
= pan_context(pctx
);
1184 unsigned new_nr
= 0;
1187 assert(start_slot
== 0);
1189 for (i
= 0; i
< num_views
; ++i
) {
1192 pipe_sampler_view_reference((struct pipe_sampler_view
**)&ctx
->sampler_views
[shader
][i
],
1196 for (; i
< ctx
->sampler_view_count
[shader
]; i
++) {
1197 pipe_sampler_view_reference((struct pipe_sampler_view
**)&ctx
->sampler_views
[shader
][i
],
1200 ctx
->sampler_view_count
[shader
] = new_nr
;
1204 panfrost_sampler_view_destroy(
1205 struct pipe_context
*pctx
,
1206 struct pipe_sampler_view
*pview
)
1208 struct panfrost_sampler_view
*view
= (struct panfrost_sampler_view
*) pview
;
1210 pipe_resource_reference(&pview
->texture
, NULL
);
1211 panfrost_bo_unreference(view
->bo
);
1216 panfrost_set_shader_buffers(
1217 struct pipe_context
*pctx
,
1218 enum pipe_shader_type shader
,
1219 unsigned start
, unsigned count
,
1220 const struct pipe_shader_buffer
*buffers
,
1221 unsigned writable_bitmask
)
1223 struct panfrost_context
*ctx
= pan_context(pctx
);
1225 util_set_shader_buffers_mask(ctx
->ssbo
[shader
], &ctx
->ssbo_mask
[shader
],
1226 buffers
, start
, count
);
1229 /* Hints that a framebuffer should use AFBC where possible */
1233 struct panfrost_screen
*screen
,
1234 const struct pipe_framebuffer_state
*fb
)
1236 /* AFBC implemenation incomplete; hide it */
1237 if (!(pan_debug
& PAN_DBG_AFBC
)) return;
1239 /* Hint AFBC to the resources bound to each color buffer */
1241 for (unsigned i
= 0; i
< fb
->nr_cbufs
; ++i
) {
1242 struct pipe_surface
*surf
= fb
->cbufs
[i
];
1243 struct panfrost_resource
*rsrc
= pan_resource(surf
->texture
);
1244 panfrost_resource_hint_layout(screen
, rsrc
, MALI_TEXTURE_AFBC
, 1);
1247 /* Also hint it to the depth buffer */
1250 struct panfrost_resource
*rsrc
= pan_resource(fb
->zsbuf
->texture
);
1251 panfrost_resource_hint_layout(screen
, rsrc
, MALI_TEXTURE_AFBC
, 1);
1256 panfrost_set_framebuffer_state(struct pipe_context
*pctx
,
1257 const struct pipe_framebuffer_state
*fb
)
1259 struct panfrost_context
*ctx
= pan_context(pctx
);
1261 panfrost_hint_afbc(pan_screen(pctx
->screen
), fb
);
1262 util_copy_framebuffer_state(&ctx
->pipe_framebuffer
, fb
);
1264 panfrost_invalidate_frame(ctx
);
1268 panfrost_create_depth_stencil_state(struct pipe_context
*pipe
,
1269 const struct pipe_depth_stencil_alpha_state
*depth_stencil
)
1271 return mem_dup(depth_stencil
, sizeof(*depth_stencil
));
1275 panfrost_bind_depth_stencil_state(struct pipe_context
*pipe
,
1278 struct panfrost_context
*ctx
= pan_context(pipe
);
1279 struct pipe_depth_stencil_alpha_state
*depth_stencil
= cso
;
1280 ctx
->depth_stencil
= depth_stencil
;
1285 /* Alpha does not exist in the hardware (it's not in ES3), so it's
1286 * emulated in the fragment shader */
1288 if (depth_stencil
->alpha
.enabled
) {
1289 /* We need to trigger a new shader (maybe) */
1290 ctx
->base
.bind_fs_state(&ctx
->base
, ctx
->shader
[PIPE_SHADER_FRAGMENT
]);
1293 /* Bounds test not implemented */
1294 assert(!depth_stencil
->depth
.bounds_test
);
1298 panfrost_delete_depth_stencil_state(struct pipe_context
*pipe
, void *depth
)
1304 panfrost_set_sample_mask(struct pipe_context
*pipe
,
1305 unsigned sample_mask
)
1310 panfrost_set_clip_state(struct pipe_context
*pipe
,
1311 const struct pipe_clip_state
*clip
)
1313 //struct panfrost_context *panfrost = pan_context(pipe);
1317 panfrost_set_viewport_states(struct pipe_context
*pipe
,
1318 unsigned start_slot
,
1319 unsigned num_viewports
,
1320 const struct pipe_viewport_state
*viewports
)
1322 struct panfrost_context
*ctx
= pan_context(pipe
);
1324 assert(start_slot
== 0);
1325 assert(num_viewports
== 1);
1327 ctx
->pipe_viewport
= *viewports
;
1331 panfrost_set_scissor_states(struct pipe_context
*pipe
,
1332 unsigned start_slot
,
1333 unsigned num_scissors
,
1334 const struct pipe_scissor_state
*scissors
)
1336 struct panfrost_context
*ctx
= pan_context(pipe
);
1338 assert(start_slot
== 0);
1339 assert(num_scissors
== 1);
1341 ctx
->scissor
= *scissors
;
1345 panfrost_set_polygon_stipple(struct pipe_context
*pipe
,
1346 const struct pipe_poly_stipple
*stipple
)
1348 //struct panfrost_context *panfrost = pan_context(pipe);
1352 panfrost_set_active_query_state(struct pipe_context
*pipe
,
1355 struct panfrost_context
*ctx
= pan_context(pipe
);
1356 ctx
->active_queries
= enable
;
1360 panfrost_destroy(struct pipe_context
*pipe
)
1362 struct panfrost_context
*panfrost
= pan_context(pipe
);
1364 if (panfrost
->blitter
)
1365 util_blitter_destroy(panfrost
->blitter
);
1367 if (panfrost
->blitter_wallpaper
)
1368 util_blitter_destroy(panfrost
->blitter_wallpaper
);
1370 util_unreference_framebuffer_state(&panfrost
->pipe_framebuffer
);
1371 u_upload_destroy(pipe
->stream_uploader
);
1376 static struct pipe_query
*
1377 panfrost_create_query(struct pipe_context
*pipe
,
1381 struct panfrost_query
*q
= rzalloc(pipe
, struct panfrost_query
);
1386 return (struct pipe_query
*) q
;
1390 panfrost_destroy_query(struct pipe_context
*pipe
, struct pipe_query
*q
)
1392 struct panfrost_query
*query
= (struct panfrost_query
*) q
;
1395 panfrost_bo_unreference(query
->bo
);
1403 panfrost_begin_query(struct pipe_context
*pipe
, struct pipe_query
*q
)
1405 struct panfrost_context
*ctx
= pan_context(pipe
);
1406 struct panfrost_query
*query
= (struct panfrost_query
*) q
;
1408 switch (query
->type
) {
1409 case PIPE_QUERY_OCCLUSION_COUNTER
:
1410 case PIPE_QUERY_OCCLUSION_PREDICATE
:
1411 case PIPE_QUERY_OCCLUSION_PREDICATE_CONSERVATIVE
:
1412 /* Allocate a bo for the query results to be stored */
1414 query
->bo
= panfrost_bo_create(
1415 pan_screen(ctx
->base
.screen
),
1416 sizeof(unsigned), 0);
1419 unsigned *result
= (unsigned *)query
->bo
->cpu
;
1420 *result
= 0; /* Default to 0 if nothing at all drawn. */
1421 ctx
->occlusion_query
= query
;
1424 /* Geometry statistics are computed in the driver. XXX: geom/tess
1427 case PIPE_QUERY_PRIMITIVES_GENERATED
:
1428 query
->start
= ctx
->prims_generated
;
1430 case PIPE_QUERY_PRIMITIVES_EMITTED
:
1431 query
->start
= ctx
->tf_prims_generated
;
1435 DBG("Skipping query %u\n", query
->type
);
1443 panfrost_end_query(struct pipe_context
*pipe
, struct pipe_query
*q
)
1445 struct panfrost_context
*ctx
= pan_context(pipe
);
1446 struct panfrost_query
*query
= (struct panfrost_query
*) q
;
1448 switch (query
->type
) {
1449 case PIPE_QUERY_OCCLUSION_COUNTER
:
1450 case PIPE_QUERY_OCCLUSION_PREDICATE
:
1451 case PIPE_QUERY_OCCLUSION_PREDICATE_CONSERVATIVE
:
1452 ctx
->occlusion_query
= NULL
;
1454 case PIPE_QUERY_PRIMITIVES_GENERATED
:
1455 query
->end
= ctx
->prims_generated
;
1457 case PIPE_QUERY_PRIMITIVES_EMITTED
:
1458 query
->end
= ctx
->tf_prims_generated
;
1466 panfrost_get_query_result(struct pipe_context
*pipe
,
1467 struct pipe_query
*q
,
1469 union pipe_query_result
*vresult
)
1471 struct panfrost_query
*query
= (struct panfrost_query
*) q
;
1472 struct panfrost_context
*ctx
= pan_context(pipe
);
1475 switch (query
->type
) {
1476 case PIPE_QUERY_OCCLUSION_COUNTER
:
1477 case PIPE_QUERY_OCCLUSION_PREDICATE
:
1478 case PIPE_QUERY_OCCLUSION_PREDICATE_CONSERVATIVE
:
1480 panfrost_flush_all_batches(ctx
, true);
1482 /* Read back the query results */
1483 unsigned *result
= (unsigned *) query
->bo
->cpu
;
1484 unsigned passed
= *result
;
1486 if (query
->type
== PIPE_QUERY_OCCLUSION_COUNTER
) {
1487 vresult
->u64
= passed
;
1489 vresult
->b
= !!passed
;
1494 case PIPE_QUERY_PRIMITIVES_GENERATED
:
1495 case PIPE_QUERY_PRIMITIVES_EMITTED
:
1496 panfrost_flush_all_batches(ctx
, true);
1497 vresult
->u64
= query
->end
- query
->start
;
1501 DBG("Skipped query get %u\n", query
->type
);
1508 static struct pipe_stream_output_target
*
1509 panfrost_create_stream_output_target(struct pipe_context
*pctx
,
1510 struct pipe_resource
*prsc
,
1511 unsigned buffer_offset
,
1512 unsigned buffer_size
)
1514 struct pipe_stream_output_target
*target
;
1516 target
= rzalloc(pctx
, struct pipe_stream_output_target
);
1521 pipe_reference_init(&target
->reference
, 1);
1522 pipe_resource_reference(&target
->buffer
, prsc
);
1524 target
->context
= pctx
;
1525 target
->buffer_offset
= buffer_offset
;
1526 target
->buffer_size
= buffer_size
;
1532 panfrost_stream_output_target_destroy(struct pipe_context
*pctx
,
1533 struct pipe_stream_output_target
*target
)
1535 pipe_resource_reference(&target
->buffer
, NULL
);
1536 ralloc_free(target
);
1540 panfrost_set_stream_output_targets(struct pipe_context
*pctx
,
1541 unsigned num_targets
,
1542 struct pipe_stream_output_target
**targets
,
1543 const unsigned *offsets
)
1545 struct panfrost_context
*ctx
= pan_context(pctx
);
1546 struct panfrost_streamout
*so
= &ctx
->streamout
;
1548 assert(num_targets
<= ARRAY_SIZE(so
->targets
));
1550 for (unsigned i
= 0; i
< num_targets
; i
++) {
1551 if (offsets
[i
] != -1)
1552 so
->offsets
[i
] = offsets
[i
];
1554 pipe_so_target_reference(&so
->targets
[i
], targets
[i
]);
1557 for (unsigned i
= 0; i
< so
->num_targets
; i
++)
1558 pipe_so_target_reference(&so
->targets
[i
], NULL
);
1560 so
->num_targets
= num_targets
;
1563 struct pipe_context
*
1564 panfrost_create_context(struct pipe_screen
*screen
, void *priv
, unsigned flags
)
1566 struct panfrost_context
*ctx
= rzalloc(screen
, struct panfrost_context
);
1567 struct pipe_context
*gallium
= (struct pipe_context
*) ctx
;
1569 gallium
->screen
= screen
;
1571 gallium
->destroy
= panfrost_destroy
;
1573 gallium
->set_framebuffer_state
= panfrost_set_framebuffer_state
;
1575 gallium
->flush
= panfrost_flush
;
1576 gallium
->clear
= panfrost_clear
;
1577 gallium
->draw_vbo
= panfrost_draw_vbo
;
1579 gallium
->set_vertex_buffers
= panfrost_set_vertex_buffers
;
1580 gallium
->set_constant_buffer
= panfrost_set_constant_buffer
;
1581 gallium
->set_shader_buffers
= panfrost_set_shader_buffers
;
1583 gallium
->set_stencil_ref
= panfrost_set_stencil_ref
;
1585 gallium
->create_sampler_view
= panfrost_create_sampler_view
;
1586 gallium
->set_sampler_views
= panfrost_set_sampler_views
;
1587 gallium
->sampler_view_destroy
= panfrost_sampler_view_destroy
;
1589 gallium
->create_rasterizer_state
= panfrost_create_rasterizer_state
;
1590 gallium
->bind_rasterizer_state
= panfrost_bind_rasterizer_state
;
1591 gallium
->delete_rasterizer_state
= panfrost_generic_cso_delete
;
1593 gallium
->create_vertex_elements_state
= panfrost_create_vertex_elements_state
;
1594 gallium
->bind_vertex_elements_state
= panfrost_bind_vertex_elements_state
;
1595 gallium
->delete_vertex_elements_state
= panfrost_generic_cso_delete
;
1597 gallium
->create_fs_state
= panfrost_create_fs_state
;
1598 gallium
->delete_fs_state
= panfrost_delete_shader_state
;
1599 gallium
->bind_fs_state
= panfrost_bind_fs_state
;
1601 gallium
->create_vs_state
= panfrost_create_vs_state
;
1602 gallium
->delete_vs_state
= panfrost_delete_shader_state
;
1603 gallium
->bind_vs_state
= panfrost_bind_vs_state
;
1605 gallium
->create_sampler_state
= panfrost_create_sampler_state
;
1606 gallium
->delete_sampler_state
= panfrost_generic_cso_delete
;
1607 gallium
->bind_sampler_states
= panfrost_bind_sampler_states
;
1609 gallium
->create_depth_stencil_alpha_state
= panfrost_create_depth_stencil_state
;
1610 gallium
->bind_depth_stencil_alpha_state
= panfrost_bind_depth_stencil_state
;
1611 gallium
->delete_depth_stencil_alpha_state
= panfrost_delete_depth_stencil_state
;
1613 gallium
->set_sample_mask
= panfrost_set_sample_mask
;
1615 gallium
->set_clip_state
= panfrost_set_clip_state
;
1616 gallium
->set_viewport_states
= panfrost_set_viewport_states
;
1617 gallium
->set_scissor_states
= panfrost_set_scissor_states
;
1618 gallium
->set_polygon_stipple
= panfrost_set_polygon_stipple
;
1619 gallium
->set_active_query_state
= panfrost_set_active_query_state
;
1621 gallium
->create_query
= panfrost_create_query
;
1622 gallium
->destroy_query
= panfrost_destroy_query
;
1623 gallium
->begin_query
= panfrost_begin_query
;
1624 gallium
->end_query
= panfrost_end_query
;
1625 gallium
->get_query_result
= panfrost_get_query_result
;
1627 gallium
->create_stream_output_target
= panfrost_create_stream_output_target
;
1628 gallium
->stream_output_target_destroy
= panfrost_stream_output_target_destroy
;
1629 gallium
->set_stream_output_targets
= panfrost_set_stream_output_targets
;
1631 panfrost_resource_context_init(gallium
);
1632 panfrost_blend_context_init(gallium
);
1633 panfrost_compute_context_init(gallium
);
1636 gallium
->stream_uploader
= u_upload_create_default(gallium
);
1637 gallium
->const_uploader
= gallium
->stream_uploader
;
1638 assert(gallium
->stream_uploader
);
1640 /* Midgard supports ES modes, plus QUADS/QUAD_STRIPS/POLYGON */
1641 ctx
->draw_modes
= (1 << (PIPE_PRIM_POLYGON
+ 1)) - 1;
1643 ctx
->primconvert
= util_primconvert_create(gallium
, ctx
->draw_modes
);
1645 ctx
->blitter
= util_blitter_create(gallium
);
1646 ctx
->blitter_wallpaper
= util_blitter_create(gallium
);
1648 assert(ctx
->blitter
);
1649 assert(ctx
->blitter_wallpaper
);
1651 /* Prepare for render! */
1653 panfrost_batch_init(ctx
);
1654 panfrost_emit_vertex_payload(ctx
);
1655 panfrost_invalidate_frame(ctx
);