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. Key by the RT and format */
  79         struct hash_table_u64 *shaders = blend->rt[rt].shaders;
  80         unsigned key = (fmt << 3) | rt;
  81
  82         struct panfrost_blend_shader *shader =
  83                 _mesa_hash_table_u64_search(shaders, key);
  84
  85         if (shader)
  86                 return shader;
  87
  88         /* Cache miss. Build one instead, cache it, and go */
  89
  90         struct panfrost_blend_shader generated =
  91                 panfrost_compile_blend_shader(ctx, &blend->base, fmt, rt);
  92
  93         shader = mem_dup(&generated, sizeof(generated));
  94         _mesa_hash_table_u64_insert(shaders, key, shader);
  95         return  shader;
  96 }
  97
  98 /* Create a blend CSO. Essentially, try to compile a fixed-function
  99  * expression and initialize blend shaders */
 100
 101 static void *
 102 panfrost_create_blend_state(struct pipe_context *pipe,
 103                             const struct pipe_blend_state *blend)
 104 {
 105         struct panfrost_context *ctx = pan_context(pipe);
 106         struct panfrost_blend_state *so = rzalloc(ctx, struct panfrost_blend_state);
 107         so->base = *blend;
 108
 109         /* TODO: The following features are not yet implemented */
 110         assert(!blend->logicop_enable);
 111         assert(!blend->alpha_to_coverage);
 112         assert(!blend->alpha_to_one);
 113
 114         for (unsigned c = 0; c < PIPE_MAX_COLOR_BUFS; ++c) {
 115                 struct panfrost_blend_rt *rt = &so->rt[c];
 116
 117                 /* There are two paths. First, we would like to try a
 118                  * fixed-function if we can */
 119
 120                 /* Without indep blending, the first RT settings replicate */
 121
 122                 unsigned g =
 123                         blend->independent_blend_enable ? c : 0;
 124
 125                 rt->has_fixed_function =
 126                         panfrost_make_fixed_blend_mode(
 127                                 &blend->rt[g],
 128                                 &rt->equation,
 129                                 &rt->constant_mask,
 130                                 blend->rt[g].colormask);
 131
 132                 /* Regardless if that works, we also need to initialize
 133                  * the blend shaders */
 134
 135                 rt->shaders = _mesa_hash_table_u64_create(so);
 136         }
 137
 138         return so;
 139 }
 140
 141 static void
 142 panfrost_bind_blend_state(struct pipe_context *pipe,
 143                           void *cso)
 144 {
 145         struct panfrost_context *ctx = pan_context(pipe);
 146         struct panfrost_screen *screen = pan_screen(ctx->base.screen);
 147         struct pipe_blend_state *blend = (struct pipe_blend_state *) cso;
 148         struct panfrost_blend_state *pblend = (struct panfrost_blend_state *) cso;
 149         ctx->blend = pblend;
 150
 151         if (!blend)
 152                 return;
 153
 154         if (screen->quirks & MIDGARD_SFBD) {
 155                 SET_BIT(ctx->fragment_shader_core.unknown2_4, MALI_NO_DITHER, !blend->dither);
 156         }
 157 }
 158
 159 static void
 160 panfrost_delete_blend_shader(struct hash_entry *entry)
 161 {
 162         struct panfrost_blend_shader *shader = (struct panfrost_blend_shader *)entry->data;
 163         free(shader->buffer);
 164         free(shader);
 165 }
 166
 167 static void
 168 panfrost_delete_blend_state(struct pipe_context *pipe,
 169                             void *cso)
 170 {
 171         struct panfrost_blend_state *blend = (struct panfrost_blend_state *) cso;
 172
 173         for (unsigned c = 0; c < 4; ++c) {
 174                 struct panfrost_blend_rt *rt = &blend->rt[c];
 175                 _mesa_hash_table_u64_clear(rt->shaders, panfrost_delete_blend_shader);
 176         }
 177         ralloc_free(blend);
 178 }
 179
 180 static void
 181 panfrost_set_blend_color(struct pipe_context *pipe,
 182                          const struct pipe_blend_color *blend_color)
 183 {
 184         struct panfrost_context *ctx = pan_context(pipe);
 185
 186         if (blend_color)
 187                 ctx->blend_color = *blend_color;
 188 }
 189
 190 /* Given a vec4 of constants, reduce it to just a single constant according to
 191  * the mask (if we can) */
 192
 193 static bool
 194 panfrost_blend_constant(float *out, float *in, unsigned mask)
 195 {
 196         /* If there is no components used, it automatically works. Do set a
 197          * dummy constant just to avoid reading uninitialized memory. */
 198
 199         if (!mask) {
 200                 *out = 0.0;
 201                 return true;
 202         }
 203
 204         /* Find some starter mask */
 205         unsigned first = ffs(mask) - 1;
 206         float cons = in[first];
 207         mask ^= (1 << first);
 208
 209         /* Ensure the rest are equal */
 210         while (mask) {
 211                 unsigned i = u_bit_scan(&mask);
 212
 213                 if (in[i] != cons) {
 214                         *out = 0.0;
 215                         return false;
 216                 }
 217         }
 218
 219         /* Otherwise, we're good to go */
 220         *out = cons;
 221         return true;
 222 }
 223
 224 /* Create a final blend given the context */
 225
 226 struct panfrost_blend_final
 227 panfrost_get_blend_for_context(struct panfrost_context *ctx, unsigned rti, struct panfrost_bo **bo, unsigned *shader_offset)
 228 {
 229         struct panfrost_batch *batch = panfrost_get_batch_for_fbo(ctx);
 230
 231         /* Grab the format, falling back gracefully if called invalidly (which
 232          * has to happen for no-color-attachment FBOs, for instance)  */
 233         struct pipe_framebuffer_state *fb = &ctx->pipe_framebuffer;
 234         enum pipe_format fmt = PIPE_FORMAT_R8G8B8A8_UNORM;
 235
 236         if ((fb->nr_cbufs > rti) && fb->cbufs[rti])
 237                 fmt = fb->cbufs[rti]->format;
 238
 239         /* Grab the blend state */
 240         struct panfrost_blend_state *blend = ctx->blend;
 241         assert(blend);
 242
 243         struct panfrost_blend_rt *rt = &blend->rt[rti];
 244
 245         struct panfrost_blend_final final;
 246
 247         /* First, we'll try a fixed function path */
 248         if (rt->has_fixed_function && panfrost_can_fixed_blend(fmt)) {
 249                 if (panfrost_blend_constant(
 250                             &final.equation.constant,
 251                             ctx->blend_color.color,
 252                             rt->constant_mask)) {
 253                         /* There's an equation and suitable constant, so we're good to go */
 254                         final.is_shader = false;
 255                         final.equation.equation = &rt->equation;
 256
 257                         final.no_blending =
 258                                 (rt->equation.rgb_mode == 0x122) &&
 259                                 (rt->equation.alpha_mode == 0x122) &&
 260                                 (rt->equation.color_mask == 0xf);
 261
 262                         return final;
 263                 }
 264         }
 265
 266         /* Otherwise, we need to grab a shader */
 267         struct panfrost_blend_shader *shader = panfrost_get_blend_shader(ctx, blend, fmt, rti);
 268         final.is_shader = true;
 269         final.no_blending = false;
 270         final.shader.work_count = shader->work_count;
 271         final.shader.first_tag = shader->first_tag;
 272
 273         /* Upload the shader, sharing a BO */
 274         if (!(*bo)) {
 275                 *bo = panfrost_batch_create_bo(batch, 4096,
 276                    PAN_BO_EXECUTE,
 277                    PAN_BO_ACCESS_PRIVATE |
 278                    PAN_BO_ACCESS_READ |
 279                    PAN_BO_ACCESS_VERTEX_TILER |
 280                    PAN_BO_ACCESS_FRAGMENT);
 281         }
 282
 283         /* Size check */
 284         assert((*shader_offset + shader->size) < 4096);
 285
 286         memcpy((*bo)->cpu + *shader_offset, shader->buffer, shader->size);
 287         final.shader.gpu = (*bo)->gpu + *shader_offset;
 288
 289         if (shader->patch_index) {
 290                 /* We have to specialize the blend shader to use constants, so
 291                  * patch in the current constants */
 292
 293                 float *patch = (float *) ((*bo)->cpu + *shader_offset + shader->patch_index);
 294                 memcpy(patch, ctx->blend_color.color, sizeof(float) * 4);
 295         }
 296
 297         *shader_offset += shader->size;
 298
 299         return final;
 300 }
 301
 302 void
 303 panfrost_blend_context_init(struct pipe_context *pipe)
 304 {
 305         pipe->create_blend_state = panfrost_create_blend_state;
 306         pipe->bind_blend_state   = panfrost_bind_blend_state;
 307         pipe->delete_blend_state = panfrost_delete_blend_state;
 308
 309         pipe->set_blend_color = panfrost_set_blend_color;
 310 }