src/mesa/drivers/dri/i965/brw_cs.c

   1 /*
   2  * Copyright (c) 2014 - 2015 Intel Corporation
   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
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  21  * DEALINGS IN THE SOFTWARE.
  22  */
  23
  24 #include "util/ralloc.h"
  25 #include "brw_context.h"
  26 #include "brw_cs.h"
  27 #include "brw_wm.h"
  28 #include "intel_mipmap_tree.h"
  29 #include "brw_state.h"
  30 #include "intel_batchbuffer.h"
  31 #include "compiler/brw_nir.h"
  32 #include "brw_program.h"
  33 #include "compiler/glsl/ir_uniform.h"
  34
  35 static void
  36 assign_cs_binding_table_offsets(const struct gen_device_info *devinfo,
  37                                 const struct gl_program *prog,
  38                                 struct brw_cs_prog_data *prog_data)
  39 {
  40    uint32_t next_binding_table_offset = 0;
  41
  42    /* May not be used if the gl_NumWorkGroups variable is not accessed. */
  43    prog_data->binding_table.work_groups_start = next_binding_table_offset;
  44    next_binding_table_offset++;
  45
  46    brw_assign_common_binding_table_offsets(devinfo, prog, &prog_data->base,
  47                                            next_binding_table_offset);
  48 }
  49
  50 static bool
  51 brw_codegen_cs_prog(struct brw_context *brw,
  52                     struct brw_program *cp,
  53                     struct brw_cs_prog_key *key)
  54 {
  55    const struct gen_device_info *devinfo = &brw->screen->devinfo;
  56    const GLuint *program;
  57    void *mem_ctx = ralloc_context(NULL);
  58    struct brw_cs_prog_data prog_data;
  59    bool start_busy = false;
  60    double start_time = 0;
  61    nir_shader *nir = nir_shader_clone(mem_ctx, cp->program.nir);
  62
  63    memset(&prog_data, 0, sizeof(prog_data));
  64
  65    if (cp->program.info.cs.shared_size > 64 * 1024) {
  66       cp->program.sh.data->LinkStatus = LINKING_FAILURE;
  67       const char *error_str =
  68          "Compute shader used more than 64KB of shared variables";
  69       ralloc_strcat(&cp->program.sh.data->InfoLog, error_str);
  70       _mesa_problem(NULL, "Failed to link compute shader: %s\n", error_str);
  71
  72       ralloc_free(mem_ctx);
  73       return false;
  74    } else {
  75       prog_data.base.total_shared = cp->program.info.cs.shared_size;
  76    }
  77
  78    assign_cs_binding_table_offsets(devinfo, &cp->program, &prog_data);
  79
  80    brw_nir_setup_glsl_uniforms(mem_ctx, nir,
  81                                &cp->program, &prog_data.base, true);
  82
  83    if (unlikely(brw->perf_debug)) {
  84       start_busy = (brw->batch.last_bo &&
  85                     brw_bo_busy(brw->batch.last_bo));
  86       start_time = get_time();
  87    }
  88
  89    int st_index = -1;
  90    if (INTEL_DEBUG & DEBUG_SHADER_TIME)
  91       st_index = brw_get_shader_time_index(brw, &cp->program, ST_CS, true);
  92
  93    char *error_str;
  94    program = brw_compile_cs(brw->screen->compiler, brw, mem_ctx, key,
  95                             &prog_data, nir, st_index, &error_str);
  96    if (program == NULL) {
  97       cp->program.sh.data->LinkStatus = LINKING_FAILURE;
  98       ralloc_strcat(&cp->program.sh.data->InfoLog, error_str);
  99       _mesa_problem(NULL, "Failed to compile compute shader: %s\n", error_str);
 100
 101       ralloc_free(mem_ctx);
 102       return false;
 103    }
 104
 105    if (unlikely(brw->perf_debug)) {
 106       if (cp->compiled_once) {
 107          brw_debug_recompile(brw, MESA_SHADER_COMPUTE, cp->program.Id,
 108                              &key->base);
 109       }
 110       cp->compiled_once = true;
 111
 112       if (start_busy && !brw_bo_busy(brw->batch.last_bo)) {
 113          perf_debug("CS compile took %.03f ms and stalled the GPU\n",
 114                     (get_time() - start_time) * 1000);
 115       }
 116    }
 117
 118    brw_alloc_stage_scratch(brw, &brw->cs.base, prog_data.base.total_scratch);
 119
 120    /* The param and pull_param arrays will be freed by the shader cache. */
 121    ralloc_steal(NULL, prog_data.base.param);
 122    ralloc_steal(NULL, prog_data.base.pull_param);
 123    brw_upload_cache(&brw->cache, BRW_CACHE_CS_PROG,
 124                     key, sizeof(*key),
 125                     program, prog_data.base.program_size,
 126                     &prog_data, sizeof(prog_data),
 127                     &brw->cs.base.prog_offset, &brw->cs.base.prog_data);
 128    ralloc_free(mem_ctx);
 129
 130    return true;
 131 }
 132
 133
 134 void
 135 brw_cs_populate_key(struct brw_context *brw, struct brw_cs_prog_key *key)
 136 {
 137    struct gl_context *ctx = &brw->ctx;
 138    /* BRW_NEW_COMPUTE_PROGRAM */
 139    const struct brw_program *cp =
 140       (struct brw_program *) brw->programs[MESA_SHADER_COMPUTE];
 141
 142    memset(key, 0, sizeof(*key));
 143
 144    /* _NEW_TEXTURE */
 145    brw_populate_base_prog_key(ctx, cp, &key->base);
 146 }
 147
 148
 149 void
 150 brw_upload_cs_prog(struct brw_context *brw)
 151 {
 152    struct gl_context *ctx = &brw->ctx;
 153    struct brw_cs_prog_key key;
 154    struct brw_program *cp =
 155       (struct brw_program *) brw->programs[MESA_SHADER_COMPUTE];
 156
 157    if (!cp)
 158       return;
 159
 160    if (!brw_state_dirty(brw, _NEW_TEXTURE, BRW_NEW_COMPUTE_PROGRAM))
 161       return;
 162
 163    brw->cs.base.sampler_count =
 164       util_last_bit(ctx->ComputeProgram._Current->SamplersUsed);
 165
 166    brw_cs_populate_key(brw, &key);
 167
 168    if (brw_search_cache(&brw->cache, BRW_CACHE_CS_PROG, &key, sizeof(key),
 169                         &brw->cs.base.prog_offset, &brw->cs.base.prog_data,
 170                         true))
 171       return;
 172
 173    if (brw_disk_cache_upload_program(brw, MESA_SHADER_COMPUTE))
 174       return;
 175
 176    cp = (struct brw_program *) brw->programs[MESA_SHADER_COMPUTE];
 177    cp->id = key.base.program_string_id;
 178
 179    MAYBE_UNUSED bool success = brw_codegen_cs_prog(brw, cp, &key);
 180    assert(success);
 181 }
 182
 183 void
 184 brw_cs_populate_default_key(const struct brw_compiler *compiler,
 185                             struct brw_cs_prog_key *key,
 186                             struct gl_program *prog)
 187 {
 188    const struct gen_device_info *devinfo = compiler->devinfo;
 189    memset(key, 0, sizeof(*key));
 190    brw_populate_default_base_prog_key(devinfo, brw_program(prog), &key->base);
 191 }
 192
 193 bool
 194 brw_cs_precompile(struct gl_context *ctx, struct gl_program *prog)
 195 {
 196    struct brw_context *brw = brw_context(ctx);
 197    struct brw_cs_prog_key key;
 198
 199    struct brw_program *bcp = brw_program(prog);
 200
 201    brw_cs_populate_default_key(brw->screen->compiler, &key, prog);
 202
 203    uint32_t old_prog_offset = brw->cs.base.prog_offset;
 204    struct brw_stage_prog_data *old_prog_data = brw->cs.base.prog_data;
 205
 206    bool success = brw_codegen_cs_prog(brw, bcp, &key);
 207
 208    brw->cs.base.prog_offset = old_prog_offset;
 209    brw->cs.base.prog_data = old_prog_data;
 210
 211    return success;
 212 }