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    }
  75
  76    assign_cs_binding_table_offsets(devinfo, &cp->program, &prog_data);
  77
  78    brw_nir_setup_glsl_uniforms(mem_ctx, nir,
  79                                &cp->program, &prog_data.base, true);
  80
  81    if (unlikely(brw->perf_debug)) {
  82       start_busy = (brw->batch.last_bo &&
  83                     brw_bo_busy(brw->batch.last_bo));
  84       start_time = get_time();
  85    }
  86
  87    int st_index = -1;
  88    if (INTEL_DEBUG & DEBUG_SHADER_TIME)
  89       st_index = brw_get_shader_time_index(brw, &cp->program, ST_CS, true);
  90
  91    char *error_str;
  92    program = brw_compile_cs(brw->screen->compiler, brw, mem_ctx, key,
  93                             &prog_data, nir, st_index, &error_str);
  94    if (program == NULL) {
  95       cp->program.sh.data->LinkStatus = LINKING_FAILURE;
  96       ralloc_strcat(&cp->program.sh.data->InfoLog, error_str);
  97       _mesa_problem(NULL, "Failed to compile compute shader: %s\n", error_str);
  98
  99       ralloc_free(mem_ctx);
 100       return false;
 101    }
 102
 103    if (unlikely(brw->perf_debug)) {
 104       if (cp->compiled_once) {
 105          brw_debug_recompile(brw, MESA_SHADER_COMPUTE, cp->program.Id,
 106                              &key->base);
 107       }
 108       cp->compiled_once = true;
 109
 110       if (start_busy && !brw_bo_busy(brw->batch.last_bo)) {
 111          perf_debug("CS compile took %.03f ms and stalled the GPU\n",
 112                     (get_time() - start_time) * 1000);
 113       }
 114    }
 115
 116    brw_alloc_stage_scratch(brw, &brw->cs.base, prog_data.base.total_scratch);
 117
 118    /* The param and pull_param arrays will be freed by the shader cache. */
 119    ralloc_steal(NULL, prog_data.base.param);
 120    ralloc_steal(NULL, prog_data.base.pull_param);
 121    brw_upload_cache(&brw->cache, BRW_CACHE_CS_PROG,
 122                     key, sizeof(*key),
 123                     program, prog_data.base.program_size,
 124                     &prog_data, sizeof(prog_data),
 125                     &brw->cs.base.prog_offset, &brw->cs.base.prog_data);
 126    ralloc_free(mem_ctx);
 127
 128    return true;
 129 }
 130
 131
 132 void
 133 brw_cs_populate_key(struct brw_context *brw, struct brw_cs_prog_key *key)
 134 {
 135    struct gl_context *ctx = &brw->ctx;
 136    /* BRW_NEW_COMPUTE_PROGRAM */
 137    const struct brw_program *cp =
 138       (struct brw_program *) brw->programs[MESA_SHADER_COMPUTE];
 139
 140    memset(key, 0, sizeof(*key));
 141
 142    /* _NEW_TEXTURE */
 143    brw_populate_base_prog_key(ctx, cp, &key->base);
 144 }
 145
 146
 147 void
 148 brw_upload_cs_prog(struct brw_context *brw)
 149 {
 150    struct gl_context *ctx = &brw->ctx;
 151    struct brw_cs_prog_key key;
 152    struct brw_program *cp =
 153       (struct brw_program *) brw->programs[MESA_SHADER_COMPUTE];
 154
 155    if (!cp)
 156       return;
 157
 158    if (!brw_state_dirty(brw, _NEW_TEXTURE, BRW_NEW_COMPUTE_PROGRAM))
 159       return;
 160
 161    brw->cs.base.sampler_count =
 162       util_last_bit(ctx->ComputeProgram._Current->SamplersUsed);
 163
 164    brw_cs_populate_key(brw, &key);
 165
 166    if (brw_search_cache(&brw->cache, BRW_CACHE_CS_PROG, &key, sizeof(key),
 167                         &brw->cs.base.prog_offset, &brw->cs.base.prog_data,
 168                         true))
 169       return;
 170
 171    if (brw_disk_cache_upload_program(brw, MESA_SHADER_COMPUTE))
 172       return;
 173
 174    cp = (struct brw_program *) brw->programs[MESA_SHADER_COMPUTE];
 175    cp->id = key.base.program_string_id;
 176
 177    ASSERTED bool success = brw_codegen_cs_prog(brw, cp, &key);
 178    assert(success);
 179 }
 180
 181 void
 182 brw_cs_populate_default_key(const struct brw_compiler *compiler,
 183                             struct brw_cs_prog_key *key,
 184                             struct gl_program *prog)
 185 {
 186    const struct gen_device_info *devinfo = compiler->devinfo;
 187    memset(key, 0, sizeof(*key));
 188    brw_populate_default_base_prog_key(devinfo, brw_program(prog), &key->base);
 189 }
 190
 191 bool
 192 brw_cs_precompile(struct gl_context *ctx, struct gl_program *prog)
 193 {
 194    struct brw_context *brw = brw_context(ctx);
 195    struct brw_cs_prog_key key;
 196
 197    struct brw_program *bcp = brw_program(prog);
 198
 199    brw_cs_populate_default_key(brw->screen->compiler, &key, prog);
 200
 201    uint32_t old_prog_offset = brw->cs.base.prog_offset;
 202    struct brw_stage_prog_data *old_prog_data = brw->cs.base.prog_data;
 203
 204    bool success = brw_codegen_cs_prog(brw, bcp, &key);
 205
 206    brw->cs.base.prog_offset = old_prog_offset;
 207    brw->cs.base.prog_data = old_prog_data;
 208
 209    return success;
 210 }