src/gallium/drivers/panfrost/pan_blend_cso.c

   1 /*
   2  * Copyright (C) 2019 Collabora, Ltd.
   3  *
   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:
  10  *
  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
  13  * Software.
  14  *
  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
  21  * SOFTWARE.
  22  *
  23  * Authors (Collabora):
  24  *   Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
  25  *
  26  */
  27
  28 #include <stdio.h>
  29 #include "util/u_memory.h"
  30 #include "pan_blend_shaders.h"
  31 #include "pan_blending.h"
  32 #include "pan_bo.h"
  33 #include "panfrost-quirks.h"
  34
  35 /* A given Gallium blend state can be encoded to the hardware in numerous,
  36  * dramatically divergent ways due to the interactions of blending with
  37  * framebuffer formats. Conceptually, there are two modes:
  38  *
  39  * - Fixed-function blending (for suitable framebuffer formats, suitable blend
  40  *   state, and suitable blend constant)
  41  *
  42  * - Blend shaders (for everything else)
  43  *
  44  * A given Gallium blend configuration will compile to exactly one
  45  * fixed-function blend state, if it compiles to any, although the constant
  46  * will vary across runs as that is tracked outside of the Gallium CSO.
  47  *
  48  * However, that same blend configuration will compile to many different blend
  49  * shaders, depending on the framebuffer formats active. The rationale is that
  50  * blend shaders override not just fixed-function blending but also
  51  * fixed-function format conversion. As such, each blend shader must be
  52  * hardcoded to a particular framebuffer format to correctly pack/unpack it. As
  53  * a concrete example, to the hardware there is no difference (!) between RG16F
  54  * and RG16UI -- both are simply 4-byte-per-pixel chunks. Thus both formats
  55  * require a blend shader (even with blending is totally disabled!), required
  56  * to do conversion as necessary (if necessary).
  57  *
  58  * All of this state is encapsulated in the panfrost_blend_state struct
  59  * (our subclass of pipe_blend_state).
  60  */
  61
  62 /* Given an initialized CSO and a particular framebuffer format, grab a
  63  * blend shader, generating and compiling it if it doesn't exist
  64  * (lazy-loading in a way). This routine, when the cache hits, should
  65  * befast, suitable for calling every draw to avoid wacky dirty
  66  * tracking paths. If the cache hits, boom, done. */
  67
  68 static struct panfrost_blend_shader *
  69 panfrost_get_blend_shader(
  70         struct panfrost_context *ctx,
  71         struct panfrost_blend_state *blend,
  72         enum pipe_format fmt,
  73         unsigned rt)
  74 {
  75         /* Prevent NULL collision issues.. */
  76         assert(fmt != 0);
  77
  78         /* Check the cache */
  79         struct hash_table_u64 *shaders = blend->rt[rt].shaders;
  80
  81         struct panfrost_blend_shader *shader =
  82                 _mesa_hash_table_u64_search(shaders, fmt);
  83
  84         if (shader)
  85                 return shader;
  86
  87         /* Cache miss. Build one instead, cache it, and go */
  88
  89         struct panfrost_blend_shader generated =
  90                 panfrost_compile_blend_shader(ctx, &blend->base, fmt);
  91
  92         shader = mem_dup(&generated, sizeof(generated));
  93         _mesa_hash_table_u64_insert(shaders, fmt, shader);
  94         return  shader;
  95 }
  96
  97 /* Create a blend CSO. Essentially, try to compile a fixed-function
  98  * expression and initialize blend shaders */
  99
 100 static void *
 101 panfrost_create_blend_state(struct pipe_context *pipe,
 102                             const struct pipe_blend_state *blend)
 103 {
 104         struct panfrost_context *ctx = pan_context(pipe);
 105         struct panfrost_blend_state *so = rzalloc(ctx, struct panfrost_blend_state);
 106         so->base = *blend;
 107
 108         /* TODO: The following features are not yet implemented */
 109         assert(!blend->logicop_enable);
 110         assert(!blend->alpha_to_coverage);
 111         assert(!blend->alpha_to_one);
 112
 113         for (unsigned c = 0; c < PIPE_MAX_COLOR_BUFS; ++c) {
 114                 struct panfrost_blend_rt *rt = &so->rt[c];
 115
 116                 /* There are two paths. First, we would like to try a
 117                  * fixed-function if we can */
 118
 119                 /* Without indep blending, the first RT settings replicate */
 120
 121                 unsigned g =
 122                         blend->independent_blend_enable ? c : 0;
 123
 124                 rt->has_fixed_function =
 125                         panfrost_make_fixed_blend_mode(
 126                                 &blend->rt[g],
 127                                 &rt->equation,
 128                                 &rt->constant_mask,
 129                                 blend->rt[g].colormask);
 130
 131                 /* Regardless if that works, we also need to initialize
 132                  * the blend shaders */
 133
 134                 rt->shaders = _mesa_hash_table_u64_create(so);
 135         }
 136
 137         return so;
 138 }
 139
 140 static void
 141 panfrost_bind_blend_state(struct pipe_context *pipe,
 142                           void *cso)
 143 {
 144         struct panfrost_context *ctx = pan_context(pipe);
 145         struct panfrost_screen *screen = pan_screen(ctx->base.screen);
 146         struct pipe_blend_state *blend = (struct pipe_blend_state *) cso;
 147         struct panfrost_blend_state *pblend = (struct panfrost_blend_state *) cso;
 148         ctx->blend = pblend;
 149
 150         if (!blend)
 151                 return;
 152
 153         if (screen->quirks & MIDGARD_SFBD) {
 154                 SET_BIT(ctx->fragment_shader_core.unknown2_4, MALI_NO_DITHER, !blend->dither);
 155         }
 156
 157         /* Shader itself is not dirty, but the shader core is */
 158         ctx->dirty |= PAN_DIRTY_FS;
 159 }
 160
 161 static void
 162 panfrost_delete_blend_shader(struct hash_entry *entry)
 163 {
 164         struct panfrost_blend_shader *shader = (struct panfrost_blend_shader *)entry->data;
 165         free(shader->buffer);
 166         free(shader);
 167 }
 168
 169 static void
 170 panfrost_delete_blend_state(struct pipe_context *pipe,
 171                             void *cso)
 172 {
 173         struct panfrost_blend_state *blend = (struct panfrost_blend_state *) cso;
 174
 175         for (unsigned c = 0; c < 4; ++c) {
 176                 struct panfrost_blend_rt *rt = &blend->rt[c];
 177                 _mesa_hash_table_u64_clear(rt->shaders, panfrost_delete_blend_shader);
 178         }
 179         ralloc_free(blend);
 180 }
 181
 182 static void
 183 panfrost_set_blend_color(struct pipe_context *pipe,
 184                          const struct pipe_blend_color *blend_color)
 185 {
 186         struct panfrost_context *ctx = pan_context(pipe);
 187
 188         if (blend_color)
 189                 ctx->blend_color = *blend_color;
 190 }
 191
 192 /* Given a vec4 of constants, reduce it to just a single constant according to
 193  * the mask (if we can) */
 194
 195 static bool
 196 panfrost_blend_constant(float *out, float *in, unsigned mask)
 197 {
 198         /* If there is no components used, it automatically works. Do set a
 199          * dummy constant just to avoid reading uninitialized memory. */
 200
 201         if (!mask) {
 202                 *out = 0.0;
 203                 return true;
 204         }
 205
 206         /* Find some starter mask */
 207         unsigned first = ffs(mask) - 1;
 208         float cons = in[first];
 209         mask ^= (1 << first);
 210
 211         /* Ensure the rest are equal */
 212         while (mask) {
 213                 unsigned i = u_bit_scan(&mask);
 214
 215                 if (in[i] != cons) {
 216                         *out = 0.0;
 217                         return false;
 218                 }
 219         }
 220
 221         /* Otherwise, we're good to go */
 222         *out = cons;
 223         return true;
 224 }
 225
 226 /* Create a final blend given the context */
 227
 228 struct panfrost_blend_final
 229 panfrost_get_blend_for_context(struct panfrost_context *ctx, unsigned rti)
 230 {
 231         struct panfrost_batch *batch = panfrost_get_batch_for_fbo(ctx);
 232
 233         /* Grab the format, falling back gracefully if called invalidly (which
 234          * has to happen for no-color-attachment FBOs, for instance)  */
 235         struct pipe_framebuffer_state *fb = &ctx->pipe_framebuffer;
 236         enum pipe_format fmt = PIPE_FORMAT_R8G8B8A8_UNORM;
 237
 238         if ((fb->nr_cbufs > rti) && fb->cbufs[rti])
 239                 fmt = fb->cbufs[rti]->format;
 240
 241         /* Grab the blend state */
 242         struct panfrost_blend_state *blend = ctx->blend;
 243         assert(blend);
 244
 245         struct panfrost_blend_rt *rt = &blend->rt[rti];
 246
 247         struct panfrost_blend_final final;
 248
 249         /* First, we'll try a fixed function path */
 250         if (rt->has_fixed_function && panfrost_can_fixed_blend(fmt)) {
 251                 if (panfrost_blend_constant(
 252                             &final.equation.constant,
 253                             ctx->blend_color.color,
 254                             rt->constant_mask)) {
 255                         /* There's an equation and suitable constant, so we're good to go */
 256                         final.is_shader = false;
 257                         final.equation.equation = &rt->equation;
 258
 259                         final.no_blending =
 260                                 (rt->equation.rgb_mode == 0x122) &&
 261                                 (rt->equation.alpha_mode == 0x122) &&
 262                                 (rt->equation.color_mask == 0xf);
 263
 264                         return final;
 265                 }
 266         }
 267
 268         /* Otherwise, we need to grab a shader */
 269         struct panfrost_blend_shader *shader = panfrost_get_blend_shader(ctx, blend, fmt, rti);
 270         final.is_shader = true;
 271         final.no_blending = false;
 272         final.shader.work_count = shader->work_count;
 273         final.shader.first_tag = shader->first_tag;
 274
 275         /* Upload the shader */
 276         final.shader.bo = panfrost_batch_create_bo(batch, shader->size,
 277                                                    PAN_BO_EXECUTE,
 278                                                    PAN_BO_ACCESS_PRIVATE |
 279                                                    PAN_BO_ACCESS_READ |
 280                                                    PAN_BO_ACCESS_VERTEX_TILER |
 281                                                    PAN_BO_ACCESS_FRAGMENT);
 282         memcpy(final.shader.bo->cpu, shader->buffer, shader->size);
 283
 284         if (shader->patch_index) {
 285                 /* We have to specialize the blend shader to use constants, so
 286                  * patch in the current constants */
 287
 288                 float *patch = (float *) (final.shader.bo->cpu + shader->patch_index);
 289                 memcpy(patch, ctx->blend_color.color, sizeof(float) * 4);
 290         }
 291
 292         return final;
 293 }
 294
 295 void
 296 panfrost_blend_context_init(struct pipe_context *pipe)
 297 {
 298         pipe->create_blend_state = panfrost_create_blend_state;
 299         pipe->bind_blend_state   = panfrost_bind_blend_state;
 300         pipe->delete_blend_state = panfrost_delete_blend_state;
 301
 302         pipe->set_blend_color = panfrost_set_blend_color;
 303 }