2 * Copyright (C) 2019 Collabora, Ltd.
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
23 * Authors (Collabora):
24 * Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
29 #include "util/u_memory.h"
30 #include "pan_blend_shaders.h"
31 #include "pan_blending.h"
33 /* A given Gallium blend state can be encoded to the hardware in numerous,
34 * dramatically divergent ways due to the interactions of blending with
35 * framebuffer formats. Conceptually, there are two modes:
37 * - Fixed-function blending (for suitable framebuffer formats, suitable blend
38 * state, and suitable blend constant)
40 * - Blend shaders (for everything else)
42 * A given Gallium blend configuration will compile to exactly one
43 * fixed-function blend state, if it compiles to any, although the constant
44 * will vary across runs as that is tracked outside of the Gallium CSO.
46 * However, that same blend configuration will compile to many different blend
47 * shaders, depending on the framebuffer formats active. The rationale is that
48 * blend shaders override not just fixed-function blending but also
49 * fixed-function format conversion. As such, each blend shader must be
50 * hardcoded to a particular framebuffer format to correctly pack/unpack it. As
51 * a concrete example, to the hardware there is no difference (!) between RG16F
52 * and RG16UI -- both are simply 4-byte-per-pixel chunks. Thus both formats
53 * require a blend shader (even with blending is totally disabled!), required
54 * to do conversion as necessary (if necessary).
56 * All of this state is encapsulated in the panfrost_blend_state struct
57 * (our subclass of pipe_blend_state).
60 /* Given an initialized CSO and a particular framebuffer format, grab a
61 * blend shader, generating and compiling it if it doesn't exist
62 * (lazy-loading in a way). This routine, when the cache hits, should
63 * befast, suitable for calling every draw to avoid wacky dirty
64 * tracking paths. If the cache hits, boom, done. */
66 static struct panfrost_blend_shader
*
67 panfrost_get_blend_shader(
68 struct panfrost_context
*ctx
,
69 struct panfrost_blend_state
*blend
,
73 /* Prevent NULL collision issues.. */
77 struct hash_table_u64
*shaders
= blend
->rt
[rt
].shaders
;
79 struct panfrost_blend_shader
*shader
=
80 _mesa_hash_table_u64_search(shaders
, fmt
);
85 /* Cache miss. Build one instead, cache it, and go */
87 struct panfrost_blend_shader generated
=
88 panfrost_compile_blend_shader(ctx
, &blend
->base
, fmt
);
90 shader
= mem_dup(&generated
, sizeof(generated
));
91 _mesa_hash_table_u64_insert(shaders
, fmt
, shader
);
95 /* Create a blend CSO. Essentially, try to compile a fixed-function
96 * expression and initialize blend shaders */
99 panfrost_create_blend_state(struct pipe_context
*pipe
,
100 const struct pipe_blend_state
*blend
)
102 struct panfrost_context
*ctx
= pan_context(pipe
);
103 struct panfrost_blend_state
*so
= rzalloc(ctx
, struct panfrost_blend_state
);
106 /* TODO: The following features are not yet implemented */
107 assert(!blend
->logicop_enable
);
108 assert(!blend
->alpha_to_coverage
);
109 assert(!blend
->alpha_to_one
);
111 for (unsigned c
= 0; c
< 4; ++c
) {
112 struct panfrost_blend_rt
*rt
= &so
->rt
[c
];
114 /* There are two paths. First, we would like to try a
115 * fixed-function if we can */
117 rt
->has_fixed_function
=
118 panfrost_make_fixed_blend_mode(
122 blend
->rt
[c
].colormask
);
124 /* Regardless if that works, we also need to initialize
125 * the blend shaders */
127 rt
->shaders
= _mesa_hash_table_u64_create(so
);
134 panfrost_bind_blend_state(struct pipe_context
*pipe
,
137 struct panfrost_context
*ctx
= pan_context(pipe
);
138 struct pipe_blend_state
*blend
= (struct pipe_blend_state
*) cso
;
139 struct panfrost_blend_state
*pblend
= (struct panfrost_blend_state
*) cso
;
145 SET_BIT(ctx
->fragment_shader_core
.unknown2_4
, MALI_NO_DITHER
, !blend
->dither
);
147 /* Shader itself is not dirty, but the shader core is */
148 ctx
->dirty
|= PAN_DIRTY_FS
;
152 panfrost_delete_blend_state(struct pipe_context
*pipe
,
155 /* TODO: Free shader binary? */
160 panfrost_set_blend_color(struct pipe_context
*pipe
,
161 const struct pipe_blend_color
*blend_color
)
163 struct panfrost_context
*ctx
= pan_context(pipe
);
166 ctx
->blend_color
= *blend_color
;
169 /* Given a vec4 of constants, reduce it to just a single constant according to
170 * the mask (if we can) */
173 panfrost_blend_constant(float *out
, float *in
, unsigned mask
)
175 /* If there is no components used, it automatically works. Do set a
176 * dummy constant just to avoid reading uninitialized memory. */
183 /* Find some starter mask */
184 unsigned first
= ffs(mask
) - 1;
185 float cons
= in
[first
];
186 mask
^= (1 << first
);
188 /* Ensure the rest are equal */
190 unsigned i
= u_bit_scan(&mask
);
198 /* Otherwise, we're good to go */
203 /* Create a final blend given the context */
205 struct panfrost_blend_final
206 panfrost_get_blend_for_context(struct panfrost_context
*ctx
, unsigned rti
)
208 /* Grab the format, falling back gracefully if called invalidly (which
209 * has to happen for no-color-attachment FBOs, for instance) */
210 struct pipe_framebuffer_state
*fb
= &ctx
->pipe_framebuffer
;
211 enum pipe_format fmt
= PIPE_FORMAT_R8G8B8A8_UNORM
;
213 if (fb
->nr_cbufs
> rti
)
214 fmt
= fb
->cbufs
[rti
]->format
;
216 /* Grab the blend state */
217 struct panfrost_blend_state
*blend
= ctx
->blend
;
220 struct panfrost_blend_rt
*rt
= &blend
->rt
[rti
];
222 struct panfrost_blend_final final
;
224 /* First, we'll try a fixed function path */
225 if (rt
->has_fixed_function
&& panfrost_can_fixed_blend(fmt
)) {
226 if (panfrost_blend_constant(
227 &final
.equation
.constant
,
228 ctx
->blend_color
.color
,
229 rt
->constant_mask
)) {
230 /* There's an equation and suitable constant, so we're good to go */
231 final
.is_shader
= false;
232 final
.equation
.equation
= &rt
->equation
;
237 /* Otherwise, we need to grab a shader */
238 struct panfrost_blend_shader
*shader
= panfrost_get_blend_shader(ctx
, blend
, fmt
, rti
);
239 final
.is_shader
= true;
240 final
.shader
.work_count
= shader
->work_count
;
242 if (shader
->patch_index
) {
243 /* We have to specialize the blend shader to use constants, so
244 * patch in the current constants and upload to transient
247 float *patch
= (float *) (shader
->shader
.cpu
+ shader
->patch_index
);
248 memcpy(patch
, ctx
->blend_color
.color
, sizeof(float) * 4);
250 final
.shader
.gpu
= panfrost_upload_transient(
251 ctx
, shader
->shader
.cpu
, shader
->size
);
253 /* No need to specialize further, use the preuploaded */
254 final
.shader
.gpu
= shader
->shader
.gpu
;
257 final
.shader
.gpu
|= shader
->first_tag
;
262 panfrost_blend_context_init(struct pipe_context
*pipe
)
264 pipe
->create_blend_state
= panfrost_create_blend_state
;
265 pipe
->bind_blend_state
= panfrost_bind_blend_state
;
266 pipe
->delete_blend_state
= panfrost_delete_blend_state
;
268 pipe
->set_blend_color
= panfrost_set_blend_color
;