src/intel/vulkan/anv_perf.c

   1 /*
   2  * Copyright © 2018 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 <assert.h>
  25 #include <stdbool.h>
  26 #include <stdint.h>
  27
  28 #include "anv_private.h"
  29
  30 #include "perf/gen_perf.h"
  31 #include "perf/gen_perf_mdapi.h"
  32
  33 struct gen_perf_config *
  34 anv_get_perf(const struct gen_device_info *devinfo, int fd)
  35 {
  36    struct gen_perf_config *perf = gen_perf_new(NULL);
  37
  38    gen_perf_init_metrics(perf, devinfo, fd, false /* pipeline statistics */);
  39
  40    /* We need DRM_I915_PERF_PROP_HOLD_PREEMPTION support, only available in
  41     * perf revision 2.
  42     */
  43    if (perf->i915_perf_version < 3)
  44       goto err;
  45
  46    return perf;
  47
  48  err:
  49    ralloc_free(perf);
  50    return NULL;
  51 }
  52
  53 void
  54 anv_device_perf_init(struct anv_device *device)
  55 {
  56    device->perf_fd = -1;
  57 }
  58
  59 static int
  60 anv_device_perf_open(struct anv_device *device, uint64_t metric_id)
  61 {
  62    uint64_t properties[DRM_I915_PERF_PROP_MAX * 2];
  63    struct drm_i915_perf_open_param param;
  64    int p = 0, stream_fd;
  65
  66    properties[p++] = DRM_I915_PERF_PROP_SAMPLE_OA;
  67    properties[p++] = true;
  68
  69    properties[p++] = DRM_I915_PERF_PROP_OA_METRICS_SET;
  70    properties[p++] = metric_id;
  71
  72    properties[p++] = DRM_I915_PERF_PROP_OA_FORMAT;
  73    properties[p++] = device->info.gen >= 8 ?
  74       I915_OA_FORMAT_A32u40_A4u32_B8_C8 :
  75       I915_OA_FORMAT_A45_B8_C8;
  76
  77    properties[p++] = DRM_I915_PERF_PROP_OA_EXPONENT;
  78    properties[p++] = 31; /* slowest sampling period */
  79
  80    properties[p++] = DRM_I915_PERF_PROP_CTX_HANDLE;
  81    properties[p++] = device->context_id;
  82
  83    properties[p++] = DRM_I915_PERF_PROP_HOLD_PREEMPTION;
  84    properties[p++] = true;
  85
  86    /* If global SSEU is available, pin it to the default. This will ensure on
  87     * Gen11 for instance we use the full EU array. Initially when perf was
  88     * enabled we would use only half on Gen11 because of functional
  89     * requirements.
  90     */
  91    if (device->physical->perf->i915_perf_version >= 4) {
  92       properties[p++] = DRM_I915_PERF_PROP_GLOBAL_SSEU;
  93       properties[p++] = (uintptr_t) &device->physical->perf->sseu;
  94    }
  95
  96    memset(&param, 0, sizeof(param));
  97    param.flags = 0;
  98    param.flags |= I915_PERF_FLAG_FD_CLOEXEC | I915_PERF_FLAG_FD_NONBLOCK;
  99    param.properties_ptr = (uintptr_t)properties;
 100    param.num_properties = p / 2;
 101
 102    stream_fd = gen_ioctl(device->fd, DRM_IOCTL_I915_PERF_OPEN, &param);
 103    return stream_fd;
 104 }
 105
 106 VkResult anv_InitializePerformanceApiINTEL(
 107     VkDevice                                    _device,
 108     const VkInitializePerformanceApiInfoINTEL*  pInitializeInfo)
 109 {
 110    ANV_FROM_HANDLE(anv_device, device, _device);
 111
 112    if (!device->physical->perf)
 113       return VK_ERROR_EXTENSION_NOT_PRESENT;
 114
 115    /* Not much to do here */
 116    return VK_SUCCESS;
 117 }
 118
 119 VkResult anv_GetPerformanceParameterINTEL(
 120     VkDevice                                    _device,
 121     VkPerformanceParameterTypeINTEL             parameter,
 122     VkPerformanceValueINTEL*                    pValue)
 123 {
 124       ANV_FROM_HANDLE(anv_device, device, _device);
 125
 126       if (!device->physical->perf)
 127          return VK_ERROR_EXTENSION_NOT_PRESENT;
 128
 129       VkResult result = VK_SUCCESS;
 130       switch (parameter) {
 131       case VK_PERFORMANCE_PARAMETER_TYPE_HW_COUNTERS_SUPPORTED_INTEL:
 132          pValue->type = VK_PERFORMANCE_VALUE_TYPE_BOOL_INTEL;
 133          pValue->data.valueBool = VK_TRUE;
 134          break;
 135
 136       case VK_PERFORMANCE_PARAMETER_TYPE_STREAM_MARKER_VALID_BITS_INTEL:
 137          pValue->type = VK_PERFORMANCE_VALUE_TYPE_UINT32_INTEL;
 138          pValue->data.value32 = 25;
 139          break;
 140
 141       default:
 142          result = VK_ERROR_FEATURE_NOT_PRESENT;
 143          break;
 144       }
 145
 146       return result;
 147 }
 148
 149 VkResult anv_CmdSetPerformanceMarkerINTEL(
 150     VkCommandBuffer                             commandBuffer,
 151     const VkPerformanceMarkerInfoINTEL*         pMarkerInfo)
 152 {
 153    ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
 154
 155    cmd_buffer->intel_perf_marker = pMarkerInfo->marker;
 156
 157    return VK_SUCCESS;
 158 }
 159
 160 VkResult anv_AcquirePerformanceConfigurationINTEL(
 161     VkDevice                                    _device,
 162     const VkPerformanceConfigurationAcquireInfoINTEL* pAcquireInfo,
 163     VkPerformanceConfigurationINTEL*            pConfiguration)
 164 {
 165    ANV_FROM_HANDLE(anv_device, device, _device);
 166
 167    struct gen_perf_registers *perf_config =
 168       gen_perf_load_configuration(device->physical->perf, device->fd,
 169                                   GEN_PERF_QUERY_GUID_MDAPI);
 170    if (!perf_config)
 171       return VK_INCOMPLETE;
 172
 173    int ret = gen_perf_store_configuration(device->physical->perf, device->fd,
 174                                           perf_config, NULL /* guid */);
 175    if (ret < 0) {
 176       ralloc_free(perf_config);
 177       return VK_INCOMPLETE;
 178    }
 179
 180    *pConfiguration = (VkPerformanceConfigurationINTEL) (uint64_t) ret;
 181
 182    return VK_SUCCESS;
 183 }
 184
 185 VkResult anv_ReleasePerformanceConfigurationINTEL(
 186     VkDevice                                    _device,
 187     VkPerformanceConfigurationINTEL             _configuration)
 188 {
 189    ANV_FROM_HANDLE(anv_device, device, _device);
 190    uint64_t config = (uint64_t) _configuration;
 191
 192    gen_ioctl(device->fd, DRM_IOCTL_I915_PERF_REMOVE_CONFIG, &config);
 193
 194    return VK_SUCCESS;
 195 }
 196
 197 VkResult anv_QueueSetPerformanceConfigurationINTEL(
 198     VkQueue                                     _queue,
 199     VkPerformanceConfigurationINTEL             _configuration)
 200 {
 201    ANV_FROM_HANDLE(anv_queue, queue, _queue);
 202    struct anv_device *device = queue->device;
 203    uint64_t configuration = (uint64_t) _configuration;
 204
 205    if (device->perf_fd < 0) {
 206       device->perf_fd = anv_device_perf_open(device, configuration);
 207       if (device->perf_fd < 0)
 208          return VK_ERROR_INITIALIZATION_FAILED;
 209    } else {
 210       int ret = gen_ioctl(device->perf_fd, I915_PERF_IOCTL_CONFIG,
 211                           (void *)(uintptr_t) _configuration);
 212       if (ret < 0)
 213          return anv_device_set_lost(device, "i915-perf config failed: %m");
 214    }
 215
 216    return VK_SUCCESS;
 217 }
 218
 219 void anv_UninitializePerformanceApiINTEL(
 220     VkDevice                                    _device)
 221 {
 222    ANV_FROM_HANDLE(anv_device, device, _device);
 223
 224    if (device->perf_fd >= 0) {
 225       close(device->perf_fd);
 226       device->perf_fd = -1;
 227    }
 228 }