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
  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:
  36  *
  37  * - Fixed-function blending (for suitable framebuffer formats, suitable blend
  38  *   state, and suitable blend constant)
  39  *
  40  * - Blend shaders (for everything else)
  41  *
  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.
  45  *
  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).
  55  *
  56  * All of this state is encapsulated in the panfrost_blend_state struct
  57  * (our subclass of pipe_blend_state).
  58  */
  59
  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. */
  65
  66 static struct panfrost_blend_shader *
  67 panfrost_get_blend_shader(
  68         struct panfrost_context *ctx,
  69         struct panfrost_blend_state *blend,
  70         enum pipe_format fmt,
  71         unsigned rt)
  72 {
  73         /* Prevent NULL collision issues.. */
  74         assert(fmt != 0);
  75
  76         /* Check the cache */
  77         struct hash_table_u64 *shaders = blend->rt[rt].shaders;
  78
  79         struct panfrost_blend_shader *shader =
  80                 _mesa_hash_table_u64_search(shaders, fmt);
  81
  82         if (shader)
  83                 return shader;
  84
  85         /* Cache miss. Build one instead, cache it, and go */
  86
  87         struct panfrost_blend_shader generated =
  88                 panfrost_compile_blend_shader(ctx, &blend->base, fmt);
  89
  90         shader = mem_dup(&generated, sizeof(generated));
  91         _mesa_hash_table_u64_insert(shaders, fmt, shader);
  92         return  shader;
  93 }
  94
  95 /* Create a blend CSO. Essentially, try to compile a fixed-function
  96  * expression and initialize blend shaders */
  97
  98 static void *
  99 panfrost_create_blend_state(struct pipe_context *pipe,
 100                             const struct pipe_blend_state *blend)
 101 {
 102         struct panfrost_context *ctx = pan_context(pipe);
 103         struct panfrost_blend_state *so = rzalloc(ctx, struct panfrost_blend_state);
 104         so->base = *blend;
 105
 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);
 110
 111         for (unsigned c = 0; c < 4; ++c) {
 112                 struct panfrost_blend_rt *rt = &so->rt[c];
 113
 114                 /* There are two paths. First, we would like to try a
 115                  * fixed-function if we can */
 116
 117                 rt->has_fixed_function =
 118                         panfrost_make_fixed_blend_mode(
 119                                 &blend->rt[c],
 120                                 &rt->equation,
 121                                 &rt->constant_mask,
 122                                 blend->rt[c].colormask);
 123
 124                 /* Regardless if that works, we also need to initialize
 125                  * the blend shaders */
 126
 127                 rt->shaders = _mesa_hash_table_u64_create(NULL);
 128         }
 129
 130         return so;
 131 }
 132
 133 static void
 134 panfrost_bind_blend_state(struct pipe_context *pipe,
 135                           void *cso)
 136 {
 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;
 140         ctx->blend = pblend;
 141
 142         if (!blend)
 143                 return;
 144
 145         SET_BIT(ctx->fragment_shader_core.unknown2_4, MALI_NO_DITHER, !blend->dither);
 146
 147         /* Shader itself is not dirty, but the shader core is */
 148         ctx->dirty |= PAN_DIRTY_FS;
 149 }
 150
 151 static void
 152 panfrost_delete_blend_state(struct pipe_context *pipe,
 153                             void *blend)
 154 {
 155         /* TODO: leaks internally? */
 156
 157         ralloc_free(blend);
 158 }
 159
 160 static void
 161 panfrost_set_blend_color(struct pipe_context *pipe,
 162                          const struct pipe_blend_color *blend_color)
 163 {
 164         struct panfrost_context *ctx = pan_context(pipe);
 165
 166         if (blend_color)
 167                 ctx->blend_color = *blend_color;
 168 }
 169
 170 /* Given a vec4 of constants, reduce it to just a single constant according to
 171  * the mask (if we can) */
 172
 173 static bool
 174 panfrost_blend_constant(float *out, float *in, unsigned mask)
 175 {
 176         /* If there is no components used, it automatically works. Do set a
 177          * dummy constant just to avoid reading uninitialized memory. */
 178
 179         if (!mask) {
 180                 *out = 0.0;
 181                 return true;
 182         }
 183
 184         /* Find some starter mask */
 185         unsigned first = ffs(mask) - 1;
 186         float cons = in[first];
 187         mask ^= (1 << first);
 188
 189         /* Ensure the rest are equal */
 190         while (mask) {
 191                 unsigned i = u_bit_scan(&mask);
 192
 193                 if (in[i] != cons) {
 194                         *out = 0.0;
 195                         return false;
 196                 }
 197         }
 198
 199         /* Otherwise, we're good to go */
 200         *out = cons;
 201         return true;
 202 }
 203
 204 /* Create a final blend given the context */
 205
 206 struct panfrost_blend_final
 207 panfrost_get_blend_for_context(struct panfrost_context *ctx, unsigned rti)
 208 {
 209         /* Grab the format */
 210         struct pipe_framebuffer_state *fb = &ctx->pipe_framebuffer;
 211         assert(fb->nr_cbufs > rti);
 212         enum pipe_format fmt = fb->cbufs[rti]->format;
 213
 214         /* Grab the blend state */
 215         struct panfrost_blend_state *blend = ctx->blend;
 216         assert(blend);
 217
 218         struct panfrost_blend_rt *rt = &blend->rt[rti];
 219
 220         struct panfrost_blend_final final;
 221
 222         /* First, we'll try a fixed function path */
 223         if (rt->has_fixed_function && panfrost_can_fixed_blend(fmt)) {
 224                 if (panfrost_blend_constant(
 225                             &final.equation.constant,
 226                             ctx->blend_color.color,
 227                             rt->constant_mask)) {
 228                         /* There's an equation and suitable constant, so we're good to go */
 229                         final.is_shader = false;
 230                         final.equation.equation = &rt->equation;
 231                         return final;
 232                 }
 233         }
 234
 235         /* Otherwise, we need to grab a shader */
 236         struct panfrost_blend_shader *shader = panfrost_get_blend_shader(ctx, blend, fmt, rti);
 237         final.is_shader = true;
 238         final.shader.work_count = shader->work_count;
 239
 240         if (shader->patch_index) {
 241                 /* We have to specialize the blend shader to use constants, so
 242                  * patch in the current constants and upload to transient
 243                  * memory */
 244
 245                 float *patch = (float *) (shader->shader.cpu + shader->patch_index);
 246                 memcpy(patch, ctx->blend_color.color, sizeof(float) * 4);
 247
 248                 final.shader.gpu = panfrost_upload_transient(
 249                                            ctx, shader->shader.cpu, shader->size);
 250         } else {
 251                 /* No need to specialize further, use the preuploaded */
 252                 final.shader.gpu = shader->shader.gpu;
 253         }
 254
 255         final.shader.gpu |= shader->first_tag;
 256         return final;
 257 }
 258
 259 void
 260 panfrost_blend_context_init(struct pipe_context *pipe)
 261 {
 262         pipe->create_blend_state = panfrost_create_blend_state;
 263         pipe->bind_blend_state   = panfrost_bind_blend_state;
 264         pipe->delete_blend_state = panfrost_delete_blend_state;
 265
 266         pipe->set_blend_color = panfrost_set_blend_color;
 267 }