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_draw.h"
29 #include "util/u_string.h"
30 #include "util/u_memory.h"
31 #include "util/u_prim.h"
32 #include "util/format/u_format.h"
33 #include "util/u_helpers.h"
35 #include "freedreno_blitter.h"
36 #include "freedreno_draw.h"
37 #include "freedreno_context.h"
38 #include "freedreno_fence.h"
39 #include "freedreno_state.h"
40 #include "freedreno_resource.h"
41 #include "freedreno_query_acc.h"
42 #include "freedreno_query_hw.h"
43 #include "freedreno_util.h"
46 resource_read(struct fd_batch
*batch
, struct pipe_resource
*prsc
)
50 fd_batch_resource_used(batch
, fd_resource(prsc
), false);
54 resource_written(struct fd_batch
*batch
, struct pipe_resource
*prsc
)
58 fd_batch_resource_used(batch
, fd_resource(prsc
), true);
62 fd_draw_vbo(struct pipe_context
*pctx
, const struct pipe_draw_info
*info
)
64 struct fd_context
*ctx
= fd_context(pctx
);
65 struct fd_batch
*batch
= fd_context_batch(ctx
);
66 struct pipe_framebuffer_state
*pfb
= &batch
->framebuffer
;
67 unsigned i
, prims
, buffers
= 0, restore_buffers
= 0;
69 /* for debugging problems with indirect draw, it is convenient
70 * to be able to emulate it, to determine if game is feeding us
73 if (info
->indirect
&& (fd_mesa_debug
& FD_DBG_NOINDR
)) {
74 util_draw_indirect(pctx
, info
);
78 if (!info
->count_from_stream_output
&& !info
->indirect
&&
79 !info
->primitive_restart
&&
80 !u_trim_pipe_prim(info
->mode
, (unsigned*)&info
->count
))
83 /* TODO: push down the region versions into the tiles */
84 if (!fd_render_condition_check(pctx
))
87 /* emulate unsupported primitives: */
88 if (!fd_supported_prim(ctx
, info
->mode
)) {
89 if (ctx
->streamout
.num_targets
> 0)
90 debug_error("stream-out with emulated prims");
91 util_primconvert_save_rasterizer_state(ctx
->primconvert
, ctx
->rasterizer
);
92 util_primconvert_draw_vbo(ctx
->primconvert
, info
);
96 fd_fence_ref(&ctx
->last_fence
, NULL
);
98 /* Upload a user index buffer. */
99 struct pipe_resource
*indexbuf
= NULL
;
100 unsigned index_offset
= 0;
101 struct pipe_draw_info new_info
;
102 if (info
->index_size
) {
103 if (info
->has_user_indices
) {
104 if (!util_upload_index_buffer(pctx
, info
, &indexbuf
, &index_offset
))
107 new_info
.index
.resource
= indexbuf
;
108 new_info
.has_user_indices
= false;
111 indexbuf
= info
->index
.resource
;
116 fd_batch_reset(batch
);
117 fd_context_all_dirty(ctx
);
120 batch
->blit
= ctx
->in_blit
;
121 batch
->back_blit
= ctx
->in_shadow
;
123 /* NOTE: needs to be before resource_written(batch->query_buf), otherwise
124 * query_buf may not be created yet.
126 fd_batch_set_stage(batch
, FD_STAGE_DRAW
);
129 * Figure out the buffers/features we need:
132 mtx_lock(&ctx
->screen
->lock
);
134 if (ctx
->dirty
& (FD_DIRTY_FRAMEBUFFER
| FD_DIRTY_ZSA
)) {
135 if (fd_depth_enabled(ctx
)) {
136 if (fd_resource(pfb
->zsbuf
->texture
)->valid
) {
137 restore_buffers
|= FD_BUFFER_DEPTH
;
139 batch
->invalidated
|= FD_BUFFER_DEPTH
;
141 batch
->gmem_reason
|= FD_GMEM_DEPTH_ENABLED
;
142 if (fd_depth_write_enabled(ctx
)) {
143 buffers
|= FD_BUFFER_DEPTH
;
144 resource_written(batch
, pfb
->zsbuf
->texture
);
146 resource_read(batch
, pfb
->zsbuf
->texture
);
150 if (fd_stencil_enabled(ctx
)) {
151 if (fd_resource(pfb
->zsbuf
->texture
)->valid
) {
152 restore_buffers
|= FD_BUFFER_STENCIL
;
154 batch
->invalidated
|= FD_BUFFER_STENCIL
;
156 batch
->gmem_reason
|= FD_GMEM_STENCIL_ENABLED
;
157 buffers
|= FD_BUFFER_STENCIL
;
158 resource_written(batch
, pfb
->zsbuf
->texture
);
162 if (fd_logicop_enabled(ctx
))
163 batch
->gmem_reason
|= FD_GMEM_LOGICOP_ENABLED
;
165 for (i
= 0; i
< pfb
->nr_cbufs
; i
++) {
166 struct pipe_resource
*surf
;
171 surf
= pfb
->cbufs
[i
]->texture
;
173 if (fd_resource(surf
)->valid
) {
174 restore_buffers
|= PIPE_CLEAR_COLOR0
<< i
;
176 batch
->invalidated
|= PIPE_CLEAR_COLOR0
<< i
;
179 buffers
|= PIPE_CLEAR_COLOR0
<< i
;
181 if (fd_blend_enabled(ctx
, i
))
182 batch
->gmem_reason
|= FD_GMEM_BLEND_ENABLED
;
184 if (ctx
->dirty
& FD_DIRTY_FRAMEBUFFER
)
185 resource_written(batch
, pfb
->cbufs
[i
]->texture
);
188 /* Mark SSBOs as being written.. we don't actually know which ones are
189 * read vs written, so just assume the worst
191 if (ctx
->dirty_shader
[PIPE_SHADER_FRAGMENT
] & FD_DIRTY_SHADER_SSBO
) {
192 foreach_bit(i
, ctx
->shaderbuf
[PIPE_SHADER_FRAGMENT
].enabled_mask
)
193 resource_written(batch
, ctx
->shaderbuf
[PIPE_SHADER_FRAGMENT
].sb
[i
].buffer
);
196 if (ctx
->dirty_shader
[PIPE_SHADER_FRAGMENT
] & FD_DIRTY_SHADER_IMAGE
) {
197 foreach_bit(i
, ctx
->shaderimg
[PIPE_SHADER_FRAGMENT
].enabled_mask
) {
198 struct pipe_image_view
*img
=
199 &ctx
->shaderimg
[PIPE_SHADER_FRAGMENT
].si
[i
];
200 if (img
->access
& PIPE_IMAGE_ACCESS_WRITE
)
201 resource_written(batch
, img
->resource
);
203 resource_read(batch
, img
->resource
);
207 if (ctx
->dirty_shader
[PIPE_SHADER_VERTEX
] & FD_DIRTY_SHADER_CONST
) {
208 foreach_bit(i
, ctx
->constbuf
[PIPE_SHADER_VERTEX
].enabled_mask
)
209 resource_read(batch
, ctx
->constbuf
[PIPE_SHADER_VERTEX
].cb
[i
].buffer
);
212 if (ctx
->dirty_shader
[PIPE_SHADER_FRAGMENT
] & FD_DIRTY_SHADER_CONST
) {
213 foreach_bit(i
, ctx
->constbuf
[PIPE_SHADER_FRAGMENT
].enabled_mask
)
214 resource_read(batch
, ctx
->constbuf
[PIPE_SHADER_FRAGMENT
].cb
[i
].buffer
);
217 /* Mark VBOs as being read */
218 if (ctx
->dirty
& FD_DIRTY_VTXBUF
) {
219 foreach_bit(i
, ctx
->vtx
.vertexbuf
.enabled_mask
) {
220 assert(!ctx
->vtx
.vertexbuf
.vb
[i
].is_user_buffer
);
221 resource_read(batch
, ctx
->vtx
.vertexbuf
.vb
[i
].buffer
.resource
);
225 /* Mark index buffer as being read */
226 resource_read(batch
, indexbuf
);
228 /* Mark indirect draw buffer as being read */
230 resource_read(batch
, info
->indirect
->buffer
);
232 /* Mark textures as being read */
233 if (ctx
->dirty_shader
[PIPE_SHADER_VERTEX
] & FD_DIRTY_SHADER_TEX
) {
234 foreach_bit(i
, ctx
->tex
[PIPE_SHADER_VERTEX
].valid_textures
)
235 resource_read(batch
, ctx
->tex
[PIPE_SHADER_VERTEX
].textures
[i
]->texture
);
238 if (ctx
->dirty_shader
[PIPE_SHADER_FRAGMENT
] & FD_DIRTY_SHADER_TEX
) {
239 foreach_bit(i
, ctx
->tex
[PIPE_SHADER_FRAGMENT
].valid_textures
)
240 resource_read(batch
, ctx
->tex
[PIPE_SHADER_FRAGMENT
].textures
[i
]->texture
);
243 /* Mark streamout buffers as being written.. */
244 if (ctx
->dirty
& FD_DIRTY_STREAMOUT
) {
245 for (i
= 0; i
< ctx
->streamout
.num_targets
; i
++)
246 if (ctx
->streamout
.targets
[i
])
247 resource_written(batch
, ctx
->streamout
.targets
[i
]->buffer
);
250 resource_written(batch
, batch
->query_buf
);
252 list_for_each_entry(struct fd_acc_query
, aq
, &ctx
->acc_active_queries
, node
)
253 resource_written(batch
, aq
->prsc
);
255 mtx_unlock(&ctx
->screen
->lock
);
259 /* Counting prims in sw doesn't work for GS and tesselation. For older
260 * gens we don't have those stages and don't have the hw counters enabled,
261 * so keep the count accurate for non-patch geometry.
263 if (info
->mode
!= PIPE_PRIM_PATCHES
)
264 prims
= u_reduced_prims_for_vertices(info
->mode
, info
->count
);
268 ctx
->stats
.draw_calls
++;
270 /* TODO prims_emitted should be clipped when the stream-out buffer is
271 * not large enough. See max_tf_vtx().. probably need to move that
272 * into common code. Although a bit more annoying since a2xx doesn't
273 * use ir3 so no common way to get at the pipe_stream_output_info
274 * which is needed for this calculation.
276 if (ctx
->streamout
.num_targets
> 0)
277 ctx
->stats
.prims_emitted
+= prims
;
278 ctx
->stats
.prims_generated
+= prims
;
280 /* any buffers that haven't been cleared yet, we need to restore: */
281 batch
->restore
|= restore_buffers
& (FD_BUFFER_ALL
& ~batch
->invalidated
);
282 /* and any buffers used, need to be resolved: */
283 batch
->resolve
|= buffers
;
285 DBG("%p: %x %ux%u num_draws=%u (%s/%s)", batch
, buffers
,
286 pfb
->width
, pfb
->height
, batch
->num_draws
,
287 util_format_short_name(pipe_surface_format(pfb
->cbufs
[0])),
288 util_format_short_name(pipe_surface_format(pfb
->zsbuf
)));
290 if (ctx
->draw_vbo(ctx
, info
, index_offset
))
291 batch
->needs_flush
= true;
293 batch
->num_vertices
+= info
->count
* info
->instance_count
;
295 for (i
= 0; i
< ctx
->streamout
.num_targets
; i
++)
296 ctx
->streamout
.offsets
[i
] += info
->count
;
298 if (fd_mesa_debug
& FD_DBG_DDRAW
)
299 fd_context_all_dirty(ctx
);
301 fd_batch_check_size(batch
);
303 if (info
== &new_info
)
304 pipe_resource_reference(&indexbuf
, NULL
);
308 fd_clear(struct pipe_context
*pctx
, unsigned buffers
,
309 const union pipe_color_union
*color
, double depth
, unsigned stencil
)
311 struct fd_context
*ctx
= fd_context(pctx
);
312 struct fd_batch
*batch
= fd_context_batch(ctx
);
313 struct pipe_framebuffer_state
*pfb
= &batch
->framebuffer
;
314 unsigned cleared_buffers
;
317 /* TODO: push down the region versions into the tiles */
318 if (!fd_render_condition_check(pctx
))
321 fd_fence_ref(&ctx
->last_fence
, NULL
);
324 fd_batch_reset(batch
);
325 fd_context_all_dirty(ctx
);
328 /* pctx->clear() is only for full-surface clears, so scissor is
329 * equivalent to having GL_SCISSOR_TEST disabled:
331 batch
->max_scissor
.minx
= 0;
332 batch
->max_scissor
.miny
= 0;
333 batch
->max_scissor
.maxx
= pfb
->width
;
334 batch
->max_scissor
.maxy
= pfb
->height
;
336 /* for bookkeeping about which buffers have been cleared (and thus
337 * can fully or partially skip mem2gmem) we need to ignore buffers
338 * that have already had a draw, in case apps do silly things like
339 * clear after draw (ie. if you only clear the color buffer, but
340 * something like alpha-test causes side effects from the draw in
341 * the depth buffer, etc)
343 cleared_buffers
= buffers
& (FD_BUFFER_ALL
& ~batch
->restore
);
344 batch
->cleared
|= buffers
;
345 batch
->invalidated
|= cleared_buffers
;
347 batch
->resolve
|= buffers
;
348 batch
->needs_flush
= true;
350 mtx_lock(&ctx
->screen
->lock
);
352 if (buffers
& PIPE_CLEAR_COLOR
)
353 for (i
= 0; i
< pfb
->nr_cbufs
; i
++)
354 if (buffers
& (PIPE_CLEAR_COLOR0
<< i
))
355 resource_written(batch
, pfb
->cbufs
[i
]->texture
);
357 if (buffers
& (PIPE_CLEAR_DEPTH
| PIPE_CLEAR_STENCIL
)) {
358 resource_written(batch
, pfb
->zsbuf
->texture
);
359 batch
->gmem_reason
|= FD_GMEM_CLEARS_DEPTH_STENCIL
;
362 resource_written(batch
, batch
->query_buf
);
364 list_for_each_entry(struct fd_acc_query
, aq
, &ctx
->acc_active_queries
, node
)
365 resource_written(batch
, aq
->prsc
);
367 mtx_unlock(&ctx
->screen
->lock
);
369 DBG("%p: %x %ux%u depth=%f, stencil=%u (%s/%s)", batch
, buffers
,
370 pfb
->width
, pfb
->height
, depth
, stencil
,
371 util_format_short_name(pipe_surface_format(pfb
->cbufs
[0])),
372 util_format_short_name(pipe_surface_format(pfb
->zsbuf
)));
374 /* if per-gen backend doesn't implement ctx->clear() generic
377 bool fallback
= true;
380 fd_batch_set_stage(batch
, FD_STAGE_CLEAR
);
382 if (ctx
->clear(ctx
, buffers
, color
, depth
, stencil
)) {
383 if (fd_mesa_debug
& FD_DBG_DCLEAR
)
384 fd_context_all_dirty(ctx
);
391 fd_blitter_clear(pctx
, buffers
, color
, depth
, stencil
);
394 fd_batch_check_size(batch
);
398 fd_clear_render_target(struct pipe_context
*pctx
, struct pipe_surface
*ps
,
399 const union pipe_color_union
*color
,
400 unsigned x
, unsigned y
, unsigned w
, unsigned h
,
401 bool render_condition_enabled
)
403 DBG("TODO: x=%u, y=%u, w=%u, h=%u", x
, y
, w
, h
);
407 fd_clear_depth_stencil(struct pipe_context
*pctx
, struct pipe_surface
*ps
,
408 unsigned buffers
, double depth
, unsigned stencil
,
409 unsigned x
, unsigned y
, unsigned w
, unsigned h
,
410 bool render_condition_enabled
)
412 DBG("TODO: buffers=%u, depth=%f, stencil=%u, x=%u, y=%u, w=%u, h=%u",
413 buffers
, depth
, stencil
, x
, y
, w
, h
);
417 fd_launch_grid(struct pipe_context
*pctx
, const struct pipe_grid_info
*info
)
419 struct fd_context
*ctx
= fd_context(pctx
);
420 struct fd_batch
*batch
, *save_batch
= NULL
;
423 batch
= fd_bc_alloc_batch(&ctx
->screen
->batch_cache
, ctx
, true);
424 fd_batch_reference(&save_batch
, ctx
->batch
);
425 fd_batch_reference(&ctx
->batch
, batch
);
426 fd_context_all_dirty(ctx
);
428 mtx_lock(&ctx
->screen
->lock
);
430 /* Mark SSBOs as being written.. we don't actually know which ones are
431 * read vs written, so just assume the worst
433 foreach_bit(i
, ctx
->shaderbuf
[PIPE_SHADER_COMPUTE
].enabled_mask
)
434 resource_written(batch
, ctx
->shaderbuf
[PIPE_SHADER_COMPUTE
].sb
[i
].buffer
);
436 foreach_bit(i
, ctx
->shaderimg
[PIPE_SHADER_COMPUTE
].enabled_mask
) {
437 struct pipe_image_view
*img
=
438 &ctx
->shaderimg
[PIPE_SHADER_COMPUTE
].si
[i
];
439 if (img
->access
& PIPE_IMAGE_ACCESS_WRITE
)
440 resource_written(batch
, img
->resource
);
442 resource_read(batch
, img
->resource
);
446 foreach_bit(i
, ctx
->constbuf
[PIPE_SHADER_COMPUTE
].enabled_mask
)
447 resource_read(batch
, ctx
->constbuf
[PIPE_SHADER_COMPUTE
].cb
[i
].buffer
);
449 /* Mark textures as being read */
450 foreach_bit(i
, ctx
->tex
[PIPE_SHADER_COMPUTE
].valid_textures
)
451 resource_read(batch
, ctx
->tex
[PIPE_SHADER_COMPUTE
].textures
[i
]->texture
);
453 /* For global buffers, we don't really know if read or written, so assume
456 foreach_bit(i
, ctx
->global_bindings
.enabled_mask
)
457 resource_written(batch
, ctx
->global_bindings
.buf
[i
]);
460 resource_read(batch
, info
->indirect
);
462 mtx_unlock(&ctx
->screen
->lock
);
464 batch
->needs_flush
= true;
465 ctx
->launch_grid(ctx
, info
);
467 fd_batch_flush(batch
, false);
469 fd_batch_reference(&ctx
->batch
, save_batch
);
470 fd_context_all_dirty(ctx
);
471 fd_batch_reference(&save_batch
, NULL
);
472 fd_batch_reference(&batch
, NULL
);
476 fd_draw_init(struct pipe_context
*pctx
)
478 pctx
->draw_vbo
= fd_draw_vbo
;
479 pctx
->clear
= fd_clear
;
480 pctx
->clear_render_target
= fd_clear_render_target
;
481 pctx
->clear_depth_stencil
= fd_clear_depth_stencil
;
483 if (has_compute(fd_screen(pctx
->screen
))) {
484 pctx
->launch_grid
= fd_launch_grid
;