src/intel/common/gen_l3_config.c

   1 /*
   2  * Copyright (c) 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 DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23
  24 #include <stdlib.h>
  25 #include <math.h>
  26
  27 #include "util/macros.h"
  28 #include "main/macros.h"
  29
  30 #include "gen_l3_config.h"
  31
  32 /**
  33  * IVB/HSW validated L3 configurations.  The first entry will be used as
  34  * default by gen7_restore_default_l3_config(), otherwise the ordering is
  35  * unimportant.
  36  */
  37 static const struct gen_l3_config ivb_l3_configs[] = {
  38    /* SLM URB ALL DC  RO  IS   C   T */
  39    {{  0, 32,  0,  0, 32,  0,  0,  0 }},
  40    {{  0, 32,  0, 16, 16,  0,  0,  0 }},
  41    {{  0, 32,  0,  4,  0,  8,  4, 16 }},
  42    {{  0, 28,  0,  8,  0,  8,  4, 16 }},
  43    {{  0, 28,  0, 16,  0,  8,  4,  8 }},
  44    {{  0, 28,  0,  8,  0, 16,  4,  8 }},
  45    {{  0, 28,  0,  0,  0, 16,  4, 16 }},
  46    {{  0, 32,  0,  0,  0, 16,  0, 16 }},
  47    {{  0, 28,  0,  4, 32,  0,  0,  0 }},
  48    {{ 16, 16,  0, 16, 16,  0,  0,  0 }},
  49    {{ 16, 16,  0,  8,  0,  8,  8,  8 }},
  50    {{ 16, 16,  0,  4,  0,  8,  4, 16 }},
  51    {{ 16, 16,  0,  4,  0, 16,  4,  8 }},
  52    {{ 16, 16,  0,  0, 32,  0,  0,  0 }},
  53    {{ 0 }}
  54 };
  55
  56 /**
  57  * VLV validated L3 configurations.  \sa ivb_l3_configs.
  58  */
  59 static const struct gen_l3_config vlv_l3_configs[] = {
  60    /* SLM URB ALL DC  RO  IS   C   T */
  61    {{  0, 64,  0,  0, 32,  0,  0,  0 }},
  62    {{  0, 80,  0,  0, 16,  0,  0,  0 }},
  63    {{  0, 80,  0,  8,  8,  0,  0,  0 }},
  64    {{  0, 64,  0, 16, 16,  0,  0,  0 }},
  65    {{  0, 60,  0,  4, 32,  0,  0,  0 }},
  66    {{ 32, 32,  0, 16, 16,  0,  0,  0 }},
  67    {{ 32, 40,  0,  8, 16,  0,  0,  0 }},
  68    {{ 32, 40,  0, 16,  8,  0,  0,  0 }},
  69    {{ 0 }}
  70 };
  71
  72 /**
  73  * BDW validated L3 configurations.  \sa ivb_l3_configs.
  74  */
  75 static const struct gen_l3_config bdw_l3_configs[] = {
  76    /* SLM URB ALL DC  RO  IS   C   T */
  77    {{  0, 48, 48,  0,  0,  0,  0,  0 }},
  78    {{  0, 48,  0, 16, 32,  0,  0,  0 }},
  79    {{  0, 32,  0, 16, 48,  0,  0,  0 }},
  80    {{  0, 32,  0,  0, 64,  0,  0,  0 }},
  81    {{  0, 32, 64,  0,  0,  0,  0,  0 }},
  82    {{ 24, 16, 48,  0,  0,  0,  0,  0 }},
  83    {{ 24, 16,  0, 16, 32,  0,  0,  0 }},
  84    {{ 24, 16,  0, 32, 16,  0,  0,  0 }},
  85    {{ 0 }}
  86 };
  87
  88 /**
  89  * CHV/SKL validated L3 configurations.  \sa ivb_l3_configs.
  90  */
  91 static const struct gen_l3_config chv_l3_configs[] = {
  92    /* SLM URB ALL DC  RO  IS   C   T */
  93    {{  0, 48, 48,  0,  0,  0,  0,  0 }},
  94    {{  0, 48,  0, 16, 32,  0,  0,  0 }},
  95    {{  0, 32,  0, 16, 48,  0,  0,  0 }},
  96    {{  0, 32,  0,  0, 64,  0,  0,  0 }},
  97    {{  0, 32, 64,  0,  0,  0,  0,  0 }},
  98    {{ 32, 16, 48,  0,  0,  0,  0,  0 }},
  99    {{ 32, 16,  0, 16, 32,  0,  0,  0 }},
 100    {{ 32, 16,  0, 32, 16,  0,  0,  0 }},
 101    {{ 0 }}
 102 };
 103
 104 /**
 105  * BXT 2x6 validated L3 configurations.  \sa ivb_l3_configs.
 106  */
 107 static const struct gen_l3_config bxt_2x6_l3_configs[] = {
 108    /* SLM URB ALL DC  RO  IS   C   T */
 109    {{  0, 32, 48,  0,  0,  0,  0,  0 }},
 110    {{  0, 32,  0,  8, 40,  0,  0,  0 }},
 111    {{  0, 32,  0, 32, 16,  0,  0,  0 }},
 112    {{ 16, 16, 48,  0,  0,  0,  0,  0 }},
 113    {{ 16, 16,  0, 40,  8,  0,  0,  0 }},
 114    {{ 16, 16,  0, 16, 32,  0,  0,  0 }},
 115    {{ 0 }}
 116 };
 117
 118 /**
 119  * Return a zero-terminated array of validated L3 configurations for the
 120  * specified device.
 121  */
 122 static const struct gen_l3_config *
 123 get_l3_configs(const struct gen_device_info *devinfo)
 124 {
 125    switch (devinfo->gen) {
 126    case 7:
 127       return (devinfo->is_baytrail ? vlv_l3_configs : ivb_l3_configs);
 128
 129    case 8:
 130       return (devinfo->is_cherryview ? chv_l3_configs : bdw_l3_configs);
 131
 132    case 9:
 133    case 10:
 134       if (devinfo->l3_banks == 1)
 135          return bxt_2x6_l3_configs;
 136       return chv_l3_configs;
 137
 138    default:
 139       unreachable("Not implemented");
 140    }
 141 }
 142
 143 /**
 144  * L1-normalize a vector of L3 partition weights.
 145  */
 146 static struct gen_l3_weights
 147 norm_l3_weights(struct gen_l3_weights w)
 148 {
 149    float sz = 0;
 150
 151    for (unsigned i = 0; i < GEN_NUM_L3P; i++)
 152       sz += w.w[i];
 153
 154    for (unsigned i = 0; i < GEN_NUM_L3P; i++)
 155       w.w[i] /= sz;
 156
 157    return w;
 158 }
 159
 160 /**
 161  * Get the relative partition weights of the specified L3 configuration.
 162  */
 163 struct gen_l3_weights
 164 gen_get_l3_config_weights(const struct gen_l3_config *cfg)
 165 {
 166    if (cfg) {
 167       struct gen_l3_weights w;
 168
 169       for (unsigned i = 0; i < GEN_NUM_L3P; i++)
 170          w.w[i] = cfg->n[i];
 171
 172       return norm_l3_weights(w);
 173    } else {
 174       const struct gen_l3_weights w = { { 0 } };
 175       return w;
 176    }
 177 }
 178
 179 /**
 180  * Distance between two L3 configurations represented as vectors of weights.
 181  * Usually just the L1 metric except when the two configurations are
 182  * considered incompatible in which case the distance will be infinite.  Note
 183  * that the compatibility condition is asymmetric -- They will be considered
 184  * incompatible whenever the reference configuration \p w0 requires SLM, DC,
 185  * or URB but \p w1 doesn't provide it.
 186  */
 187 float
 188 gen_diff_l3_weights(struct gen_l3_weights w0, struct gen_l3_weights w1)
 189 {
 190    if ((w0.w[GEN_L3P_SLM] && !w1.w[GEN_L3P_SLM]) ||
 191        (w0.w[GEN_L3P_DC] && !w1.w[GEN_L3P_DC] && !w1.w[GEN_L3P_ALL]) ||
 192        (w0.w[GEN_L3P_URB] && !w1.w[GEN_L3P_URB])) {
 193       return HUGE_VALF;
 194
 195    } else {
 196       float dw = 0;
 197
 198       for (unsigned i = 0; i < GEN_NUM_L3P; i++)
 199          dw += fabs(w0.w[i] - w1.w[i]);
 200
 201       return dw;
 202    }
 203 }
 204
 205 /**
 206  * Return a reasonable default L3 configuration for the specified device based
 207  * on whether SLM and DC are required.  In the non-SLM non-DC case the result
 208  * is intended to approximately resemble the hardware defaults.
 209  */
 210 struct gen_l3_weights
 211 gen_get_default_l3_weights(const struct gen_device_info *devinfo,
 212                            bool needs_dc, bool needs_slm)
 213 {
 214    struct gen_l3_weights w = {{ 0 }};
 215
 216    w.w[GEN_L3P_SLM] = needs_slm;
 217    w.w[GEN_L3P_URB] = 1.0;
 218
 219    if (devinfo->gen >= 8) {
 220       w.w[GEN_L3P_ALL] = 1.0;
 221    } else {
 222       w.w[GEN_L3P_DC] = needs_dc ? 0.1 : 0;
 223       w.w[GEN_L3P_RO] = devinfo->is_baytrail ? 0.5 : 1.0;
 224    }
 225
 226    return norm_l3_weights(w);
 227 }
 228
 229 /**
 230  * Get the default L3 configuration
 231  */
 232 const struct gen_l3_config *
 233 gen_get_default_l3_config(const struct gen_device_info *devinfo)
 234 {
 235    /* For efficiency assume that the first entry of the array matches the
 236     * default configuration.
 237     */
 238    const struct gen_l3_config *const cfg = get_l3_configs(devinfo);
 239    assert(cfg == gen_get_l3_config(devinfo,
 240                     gen_get_default_l3_weights(devinfo, false, false)));
 241    return cfg;
 242 }
 243
 244 /**
 245  * Return the closest validated L3 configuration for the specified device and
 246  * weight vector.
 247  */
 248 const struct gen_l3_config *
 249 gen_get_l3_config(const struct gen_device_info *devinfo,
 250                   struct gen_l3_weights w0)
 251 {
 252    const struct gen_l3_config *const cfgs = get_l3_configs(devinfo);
 253    const struct gen_l3_config *cfg_best = NULL;
 254    float dw_best = HUGE_VALF;
 255
 256    for (const struct gen_l3_config *cfg = cfgs; cfg->n[GEN_L3P_URB]; cfg++) {
 257       const float dw = gen_diff_l3_weights(w0, gen_get_l3_config_weights(cfg));
 258
 259       if (dw < dw_best) {
 260          cfg_best = cfg;
 261          dw_best = dw;
 262       }
 263    }
 264
 265    return cfg_best;
 266 }
 267
 268 /**
 269  * Return the size of an L3 way in KB.
 270  */
 271 static unsigned
 272 get_l3_way_size(const struct gen_device_info *devinfo)
 273 {
 274    assert(devinfo->l3_banks);
 275
 276    if (devinfo->is_broxton)
 277       return 4;
 278
 279    return 2 * devinfo->l3_banks;
 280 }
 281
 282 /**
 283  * Return the unit brw_context::urb::size is expressed in, in KB.  \sa
 284  * gen_device_info::urb::size.
 285  */
 286 static unsigned
 287 get_urb_size_scale(const struct gen_device_info *devinfo)
 288 {
 289    return (devinfo->gen >= 8 ? devinfo->num_slices : 1);
 290 }
 291
 292 unsigned
 293 gen_get_l3_config_urb_size(const struct gen_device_info *devinfo,
 294                            const struct gen_l3_config *cfg)
 295 {
 296    /* From the SKL "L3 Allocation and Programming" documentation:
 297     *
 298     * "URB is limited to 1008KB due to programming restrictions.  This is not
 299     * a restriction of the L3 implementation, but of the FF and other clients.
 300     * Therefore, in a GT4 implementation it is possible for the programmed
 301     * allocation of the L3 data array to provide 3*384KB=1152KB for URB, but
 302     * only 1008KB of this will be used."
 303     */
 304    const unsigned max = (devinfo->gen == 9 ? 1008 : ~0);
 305    return MIN2(max, cfg->n[GEN_L3P_URB] * get_l3_way_size(devinfo)) /
 306           get_urb_size_scale(devinfo);
 307 }
 308
 309 /**
 310  * Print out the specified L3 configuration.
 311  */
 312 void
 313 gen_dump_l3_config(const struct gen_l3_config *cfg, FILE *fp)
 314 {
 315    fprintf(stderr, "SLM=%d URB=%d ALL=%d DC=%d RO=%d IS=%d C=%d T=%d\n",
 316            cfg->n[GEN_L3P_SLM], cfg->n[GEN_L3P_URB], cfg->n[GEN_L3P_ALL],
 317            cfg->n[GEN_L3P_DC], cfg->n[GEN_L3P_RO],
 318            cfg->n[GEN_L3P_IS], cfg->n[GEN_L3P_C], cfg->n[GEN_L3P_T]);
 319 }