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->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                 if (!blend->logicop_enable) {
 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
 133                 /* Regardless if that works, we also need to initialize
 134                  * the blend shaders */
 135
 136                 rt->shaders = _mesa_hash_table_u64_create(so);
 137         }
 138
 139         return so;
 140 }
 141
 142 static void
 143 panfrost_bind_blend_state(struct pipe_context *pipe,
 144                           void *cso)
 145 {
 146         struct panfrost_context *ctx = pan_context(pipe);
 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
 155 static void
 156 panfrost_delete_blend_shader(struct hash_entry *entry)
 157 {
 158         struct panfrost_blend_shader *shader = (struct panfrost_blend_shader *)entry->data;
 159         free(shader->buffer);
 160         free(shader);
 161 }
 162
 163 static void
 164 panfrost_delete_blend_state(struct pipe_context *pipe,
 165                             void *cso)
 166 {
 167         struct panfrost_blend_state *blend = (struct panfrost_blend_state *) cso;
 168
 169         for (unsigned c = 0; c < 4; ++c) {
 170                 struct panfrost_blend_rt *rt = &blend->rt[c];
 171                 _mesa_hash_table_u64_clear(rt->shaders, panfrost_delete_blend_shader);
 172         }
 173         ralloc_free(blend);
 174 }
 175
 176 static void
 177 panfrost_set_blend_color(struct pipe_context *pipe,
 178                          const struct pipe_blend_color *blend_color)
 179 {
 180         struct panfrost_context *ctx = pan_context(pipe);
 181
 182         if (blend_color)
 183                 ctx->blend_color = *blend_color;
 184 }
 185
 186 /* Given a vec4 of constants, reduce it to just a single constant according to
 187  * the mask (if we can) */
 188
 189 static bool
 190 panfrost_blend_constant(float *out, float *in, unsigned mask)
 191 {
 192         /* If there is no components used, it automatically works. Do set a
 193          * dummy constant just to avoid reading uninitialized memory. */
 194
 195         if (!mask) {
 196                 *out = 0.0;
 197                 return true;
 198         }
 199
 200         /* Find some starter mask */
 201         unsigned first = ffs(mask) - 1;
 202         float cons = in[first];
 203         mask ^= (1 << first);
 204
 205         /* Ensure the rest are equal */
 206         while (mask) {
 207                 unsigned i = u_bit_scan(&mask);
 208
 209                 if (in[i] != cons) {
 210                         *out = 0.0;
 211                         return false;
 212                 }
 213         }
 214
 215         /* Otherwise, we're good to go */
 216         *out = cons;
 217         return true;
 218 }
 219
 220 /* Create a final blend given the context */
 221
 222 struct panfrost_blend_final
 223 panfrost_get_blend_for_context(struct panfrost_context *ctx, unsigned rti, struct panfrost_bo **bo, unsigned *shader_offset)
 224 {
 225         struct panfrost_batch *batch = panfrost_get_batch_for_fbo(ctx);
 226
 227         /* Grab the format, falling back gracefully if called invalidly (which
 228          * has to happen for no-color-attachment FBOs, for instance)  */
 229         struct pipe_framebuffer_state *fb = &ctx->pipe_framebuffer;
 230         enum pipe_format fmt = PIPE_FORMAT_R8G8B8A8_UNORM;
 231
 232         if ((fb->nr_cbufs > rti) && fb->cbufs[rti])
 233                 fmt = fb->cbufs[rti]->format;
 234
 235         /* Grab the blend state */
 236         struct panfrost_blend_state *blend = ctx->blend;
 237         assert(blend);
 238
 239         struct panfrost_blend_rt *rt = &blend->rt[rti];
 240
 241         struct panfrost_blend_final final;
 242
 243         /* First, we'll try a fixed function path */
 244         if (rt->has_fixed_function && panfrost_can_fixed_blend(fmt)) {
 245                 if (panfrost_blend_constant(
 246                             &final.equation.constant,
 247                             ctx->blend_color.color,
 248                             rt->constant_mask)) {
 249                         /* There's an equation and suitable constant, so we're good to go */
 250                         final.is_shader = false;
 251                         final.equation.equation = &rt->equation;
 252
 253                         final.no_blending =
 254                                 (rt->equation.rgb_mode == 0x122) &&
 255                                 (rt->equation.alpha_mode == 0x122) &&
 256                                 (rt->equation.color_mask == 0xf);
 257
 258                         return final;
 259                 }
 260         }
 261
 262         /* Otherwise, we need to grab a shader */
 263         struct panfrost_blend_shader *shader = panfrost_get_blend_shader(ctx, blend, fmt, rti);
 264         final.is_shader = true;
 265         final.no_blending = false;
 266         final.shader.work_count = shader->work_count;
 267         final.shader.first_tag = shader->first_tag;
 268
 269         /* Upload the shader, sharing a BO */
 270         if (!(*bo)) {
 271                 *bo = panfrost_batch_create_bo(batch, 4096,
 272                    PAN_BO_EXECUTE,
 273                    PAN_BO_ACCESS_PRIVATE |
 274                    PAN_BO_ACCESS_READ |
 275                    PAN_BO_ACCESS_VERTEX_TILER |
 276                    PAN_BO_ACCESS_FRAGMENT);
 277         }
 278
 279         /* Size check */
 280         assert((*shader_offset + shader->size) < 4096);
 281
 282         memcpy((*bo)->cpu + *shader_offset, shader->buffer, shader->size);
 283         final.shader.gpu = (*bo)->gpu + *shader_offset;
 284
 285         if (shader->patch_index) {
 286                 /* We have to specialize the blend shader to use constants, so
 287                  * patch in the current constants */
 288
 289                 float *patch = (float *) ((*bo)->cpu + *shader_offset + shader->patch_index);
 290                 memcpy(patch, ctx->blend_color.color, sizeof(float) * 4);
 291         }
 292
 293         *shader_offset += shader->size;
 294
 295         return final;
 296 }
 297
 298 void
 299 panfrost_blend_context_init(struct pipe_context *pipe)
 300 {
 301         pipe->create_blend_state = panfrost_create_blend_state;
 302         pipe->bind_blend_state   = panfrost_bind_blend_state;
 303         pipe->delete_blend_state = panfrost_delete_blend_state;
 304
 305         pipe->set_blend_color = panfrost_set_blend_color;
 306 }