intel/perf: add TGL support

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>

src/intel/perf/gen_perf.c

@@ -789,6 +789,8 @@ get_register_queries_function(const struct gen_device_info *devinfo)
       return gen_oa_register_queries_cnl;
    if (devinfo->gen == 11)
       return gen_oa_register_queries_icl;
+   if (devinfo->gen == 12)
+      return gen_oa_register_queries_tgl;
 
    return NULL;
 }
@@ -2254,6 +2256,14 @@ accumulate_oa_reports(struct gen_perf_context *perf_ctx,
       goto error;
    }
 
+   /* On Gen12+ OA reports are sourced from per context counters, so we don't
+    * ever have to look at the global OA buffer. Yey \o/
+    */
+   if (perf_ctx->devinfo->gen >= 12) {
+      last = start;
+      goto end;
+   }
+
    /* See if we have any periodic reports to accumulate too... */
 
    /* N.B. The oa.samples_head was set when the query began and

src/intel/perf/gen_perf.py

@@ -181,7 +181,10 @@ hw_vars["$EuSlicesTotalCount"] = "perf->sys_vars.n_eu_slices"
 hw_vars["$EuSubslicesTotalCount"] = "perf->sys_vars.n_eu_sub_slices"
 hw_vars["$EuThreadsCount"] = "perf->sys_vars.eu_threads_count"
 hw_vars["$SliceMask"] = "perf->sys_vars.slice_mask"
+# subslice_mask is interchangeable with subslice/dual-subslice since Gen12+
+# only has dual subslices which can be assimilated with 16EUs subslices.
 hw_vars["$SubsliceMask"] = "perf->sys_vars.subslice_mask"
+hw_vars["$DualSubsliceMask"] = "perf->sys_vars.subslice_mask"
 hw_vars["$GpuTimestampFrequency"] = "perf->sys_vars.timestamp_frequency"
 hw_vars["$GpuMinFrequency"] = "perf->sys_vars.gt_min_freq"
 hw_vars["$GpuMaxFrequency"] = "perf->sys_vars.gt_max_freq"

src/intel/perf/meson.build

@@ -7,6 +7,7 @@ gen_hw_metrics = [
   'bxt', 'glk',
   'cnl',
   'icl',
+  'tgl',
 ]
 
 gen_hw_metrics_xml_files = []