src/gallium/drivers/freedreno/ir3/ir3_gallium.c

   1 /*
   2  * Copyright (C) 2014 Rob Clark <robclark@freedesktop.org>
   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:
  24  *    Rob Clark <robclark@freedesktop.org>
  25  */
  26
  27 #include "pipe/p_state.h"
  28 #include "pipe/p_screen.h"
  29 #include "util/u_string.h"
  30 #include "util/u_memory.h"
  31 #include "util/u_inlines.h"
  32 #include "util/format/u_format.h"
  33 #include "tgsi/tgsi_dump.h"
  34 #include "tgsi/tgsi_parse.h"
  35
  36 #include "nir/tgsi_to_nir.h"
  37
  38 #include "freedreno_context.h"
  39 #include "freedreno_util.h"
  40
  41 #include "ir3/ir3_shader.h"
  42 #include "ir3/ir3_gallium.h"
  43 #include "ir3/ir3_compiler.h"
  44 #include "ir3/ir3_nir.h"
  45
  46 static void
  47 dump_shader_info(struct ir3_shader_variant *v, bool binning_pass,
  48                 struct pipe_debug_callback *debug)
  49 {
  50         if (!unlikely(fd_mesa_debug & FD_DBG_SHADERDB))
  51                 return;
  52
  53         pipe_debug_message(debug, SHADER_INFO,
  54                         "%s shader: %u inst, %u nops, %u non-nops, %u mov, %u cov, "
  55                         "%u dwords, %u last-baryf, %u half, %u full, %u constlen, "
  56                         "%u sstall, %u (ss), %u (sy), %d max_sun, %d loops\n",
  57                         ir3_shader_stage(v),
  58                         v->info.instrs_count,
  59                         v->info.nops_count,
  60                         v->info.instrs_count - v->info.nops_count,
  61                         v->info.mov_count,
  62                         v->info.cov_count,
  63                         v->info.sizedwords,
  64                         v->info.last_baryf,
  65                         v->info.max_half_reg + 1,
  66                         v->info.max_reg + 1,
  67                         v->constlen,
  68                         v->info.sstall,
  69                         v->info.ss, v->info.sy,
  70                         v->max_sun, v->loops);
  71 }
  72
  73 struct ir3_shader_variant *
  74 ir3_shader_variant(struct ir3_shader *shader, struct ir3_shader_key key,
  75                 bool binning_pass, struct pipe_debug_callback *debug)
  76 {
  77         struct ir3_shader_variant *v;
  78         bool created = false;
  79
  80         /* Some shader key values may not be used by a given ir3_shader (for
  81          * example, fragment shader saturates in the vertex shader), so clean out
  82          * those flags to avoid recompiling.
  83          */
  84         ir3_key_clear_unused(&key, shader);
  85
  86         v = ir3_shader_get_variant(shader, &key, binning_pass, &created);
  87
  88         if (created) {
  89                 if (shader->initial_variants_done) {
  90                         pipe_debug_message(debug, SHADER_INFO,
  91                                         "%s shader: recompiling at draw time: global 0x%08x, vsats %x/%x/%x, fsats %x/%x/%x, vfsamples %x/%x, astc %x/%x\n",
  92                                         ir3_shader_stage(v),
  93                                         key.global,
  94                                         key.vsaturate_s, key.vsaturate_t, key.vsaturate_r,
  95                                         key.fsaturate_s, key.fsaturate_t, key.fsaturate_r,
  96                                         key.vsamples, key.fsamples,
  97                                         key.vastc_srgb, key.fastc_srgb);
  98
  99                 }
 100                 dump_shader_info(v, binning_pass, debug);
 101         }
 102
 103         return v;
 104 }
 105
 106 static void
 107 copy_stream_out(struct ir3_stream_output_info *i,
 108                 const struct pipe_stream_output_info *p)
 109 {
 110         STATIC_ASSERT(ARRAY_SIZE(i->stride) == ARRAY_SIZE(p->stride));
 111         STATIC_ASSERT(ARRAY_SIZE(i->output) == ARRAY_SIZE(p->output));
 112
 113         i->num_outputs = p->num_outputs;
 114         for (int n = 0; n < ARRAY_SIZE(i->stride); n++)
 115                 i->stride[n] = p->stride[n];
 116
 117         for (int n = 0; n < ARRAY_SIZE(i->output); n++) {
 118                 i->output[n].register_index  = p->output[n].register_index;
 119                 i->output[n].start_component = p->output[n].start_component;
 120                 i->output[n].num_components  = p->output[n].num_components;
 121                 i->output[n].output_buffer   = p->output[n].output_buffer;
 122                 i->output[n].dst_offset      = p->output[n].dst_offset;
 123                 i->output[n].stream          = p->output[n].stream;
 124         }
 125 }
 126
 127 struct ir3_shader *
 128 ir3_shader_create(struct ir3_compiler *compiler,
 129                 const struct pipe_shader_state *cso,
 130                 struct pipe_debug_callback *debug,
 131                 struct pipe_screen *screen)
 132 {
 133         nir_shader *nir;
 134         if (cso->type == PIPE_SHADER_IR_NIR) {
 135                 /* we take ownership of the reference: */
 136                 nir = cso->ir.nir;
 137         } else {
 138                 debug_assert(cso->type == PIPE_SHADER_IR_TGSI);
 139                 if (ir3_shader_debug & IR3_DBG_DISASM) {
 140                         tgsi_dump(cso->tokens, 0);
 141                 }
 142                 nir = tgsi_to_nir(cso->tokens, screen, false);
 143         }
 144
 145         struct ir3_stream_output_info stream_output;
 146         copy_stream_out(&stream_output, &cso->stream_output);
 147
 148         struct ir3_shader *shader = ir3_shader_from_nir(compiler, nir, &stream_output);
 149
 150         /* Compile standard variants immediately to try to avoid draw-time stalls
 151          * to run the compiler.
 152          */
 153         struct ir3_shader_key key = {
 154                 .tessellation = IR3_TESS_NONE,
 155                 .msaa = true,
 156         };
 157
 158         switch (nir->info.stage) {
 159         case MESA_SHADER_TESS_EVAL:
 160                 key.tessellation = ir3_tess_mode(nir->info.tess.primitive_mode);
 161                 break;
 162
 163         case MESA_SHADER_TESS_CTRL:
 164                 /* The primitive_mode field, while it exists for TCS, is not
 165                  * populated (since separable shaders between TCS/TES are legal,
 166                  * so TCS wouldn't have access to TES's declaration).  Make a
 167                  * guess so that we shader-db something plausible for TCS.
 168                  */
 169                 if (nir->info.outputs_written & VARYING_BIT_TESS_LEVEL_INNER)
 170                         key.tessellation = IR3_TESS_TRIANGLES;
 171                 else
 172                         key.tessellation = IR3_TESS_ISOLINES;
 173                 break;
 174
 175         case MESA_SHADER_GEOMETRY:
 176                 key.has_gs = true;
 177                 break;
 178
 179         default:
 180                 break;
 181         }
 182
 183         ir3_shader_variant(shader, key, false, debug);
 184
 185         if (nir->info.stage == MESA_SHADER_VERTEX)
 186                 ir3_shader_variant(shader, key, true, debug);
 187
 188         shader->initial_variants_done = true;
 189
 190         return shader;
 191 }
 192
 193 /* a bit annoying that compute-shader and normal shader state objects
 194  * aren't a bit more aligned.
 195  */
 196 struct ir3_shader *
 197 ir3_shader_create_compute(struct ir3_compiler *compiler,
 198                 const struct pipe_compute_state *cso,
 199                 struct pipe_debug_callback *debug,
 200                 struct pipe_screen *screen)
 201 {
 202         nir_shader *nir;
 203         if (cso->ir_type == PIPE_SHADER_IR_NIR) {
 204                 /* we take ownership of the reference: */
 205                 nir = (nir_shader *)cso->prog;
 206         } else {
 207                 debug_assert(cso->ir_type == PIPE_SHADER_IR_TGSI);
 208                 if (ir3_shader_debug & IR3_DBG_DISASM) {
 209                         tgsi_dump(cso->prog, 0);
 210                 }
 211                 nir = tgsi_to_nir(cso->prog, screen, false);
 212         }
 213
 214         struct ir3_shader *shader = ir3_shader_from_nir(compiler, nir, NULL);
 215
 216         /* Immediately compile a standard variant.  We have so few variants in our
 217          * shaders, that doing so almost eliminates draw-time recompiles.  (This
 218          * is also how we get data from shader-db's ./run)
 219          */
 220         static struct ir3_shader_key key; /* static is implicitly zeroed */
 221         ir3_shader_variant(shader, key, false, debug);
 222
 223         shader->initial_variants_done = true;
 224
 225         return shader;
 226 }
 227
 228 static void *
 229 ir3_shader_state_create(struct pipe_context *pctx, const struct pipe_shader_state *cso)
 230 {
 231         struct fd_context *ctx = fd_context(pctx);
 232         struct ir3_compiler *compiler = ctx->screen->compiler;
 233         return ir3_shader_create(compiler, cso, &ctx->debug, pctx->screen);
 234 }
 235
 236 static void
 237 ir3_shader_state_delete(struct pipe_context *pctx, void *hwcso)
 238 {
 239         struct ir3_shader *so = hwcso;
 240         ir3_shader_destroy(so);
 241 }
 242
 243 void
 244 ir3_prog_init(struct pipe_context *pctx)
 245 {
 246         pctx->create_vs_state = ir3_shader_state_create;
 247         pctx->delete_vs_state = ir3_shader_state_delete;
 248
 249         pctx->create_tcs_state = ir3_shader_state_create;
 250         pctx->delete_tcs_state = ir3_shader_state_delete;
 251
 252         pctx->create_tes_state = ir3_shader_state_create;
 253         pctx->delete_tes_state = ir3_shader_state_delete;
 254
 255         pctx->create_gs_state = ir3_shader_state_create;
 256         pctx->delete_gs_state = ir3_shader_state_delete;
 257
 258         pctx->create_fs_state = ir3_shader_state_create;
 259         pctx->delete_fs_state = ir3_shader_state_delete;
 260 }